fteqw/engine/client/image.c

#include "quakedef.h"
#include "shader.h"
#include "glquake.h"	//we need some of the gl format enums

//#define PURGEIMAGES	//somewhat experimental still. we're still flushing more than we should.

//FIXME
texid_t GL_FindTextureFallback (const char *identifier, unsigned int flags, void *fallback, int fallbackwidth, int fallbackheight, uploadfmt_t fallbackfmt);
//FIXME

#ifdef NPFTE
//#define Con_Printf(f, ...)
//hope you're on a littleendian machine
#define LittleShort(s) s
#define LittleLong(s) s
#else
cvar_t r_dodgytgafiles = CVARD("r_dodgytgafiles", "0", "Many old glquake engines had a buggy tga loader that ignored bottom-up flags. Naturally people worked around this and the world was plagued with buggy images. Most engines have now fixed the bug, but you can reenable it if you have bugged tga files.");
cvar_t r_dodgypcxfiles = CVARD("r_dodgypcxfiles", "0", "When enabled, this will ignore the palette stored within pcx files, for compatibility with quake2.");
cvar_t r_dodgymiptex = CVARD("r_dodgymiptex", "1", "When enabled, this will force regeneration of mipmaps, discarding mips1-4 like glquake did. This may eg solve fullbright issues with some maps, but may reduce distant detail levels.");

char *r_defaultimageextensions =
#ifdef IMAGEFMT_DDS
	"dds "	//compressed or something
#endif
#ifdef IMAGEFMT_KTX
	"ktx "	//compressed or something. not to be confused with the qw mod by the same name. GL requires that etc2 compression is supported by modern drivers, but not necessarily the hardware. as such, dds with its s3tc bias should always come first (as the patents mean that drivers are much less likely to advertise it when they don't support it properly).
#endif
	"tga"	//fairly fast to load
#if defined(AVAIL_PNGLIB) || defined(FTE_TARGET_WEB)
	" png"	//pngs, fairly common, but slow
#endif
#ifdef IMAGEFMT_BMP
	//" bmp"	//wtf? at least not lossy
	//" ico"	//noone wants this...
#endif
#if defined(AVAIL_JPEGLIB) || defined(FTE_TARGET_WEB)
	" jpg"	//q3 uses some jpegs, for some reason
	//" jpeg"	//thankfuly the quake community stuck to .jpg instead
#endif
#ifdef IMAGEFMT_PBM
	//" pfm" //float version (technically seperate, but similarish)
	//" pbm" //1-bit v grey
	//" pgm" //greyscale
	//" ppm" //rgb values
	//" pam" //'arbitrary' version
#endif
#ifdef IMAGEFMT_PSD
	//" psd" //paintshop images (8bit+16bit, but base layer only)
#endif
#ifdef IMAGEFMT_HDR
	//" hdr" //some file that uses RGBE formatted data, for hdr images.
#endif
#ifdef IMAGEFMT_PKM
	//" pkm"	//compressed format, but lacks mipmaps which makes it terrible to use.
#endif
#ifdef IMAGEFMT_PCX
	" pcx"	//pcxes are the original gamedata of q2. So we don't want them to override pngs.
#endif
	;
static void Image_ChangeFormat(struct pendingtextureinfo *mips, unsigned int flags, uploadfmt_t origfmt);
static void QDECL R_ImageExtensions_Callback(struct cvar_s *var, char *oldvalue);
cvar_t r_imageextensions			= CVARCD("r_imageextensions", NULL, R_ImageExtensions_Callback, "The list of image file extensions which might exist on disk (note that this does not list all supported formats, only the extensions that should be searched for).");
cvar_t r_image_downloadsizelimit	= CVARFD("r_image_downloadsizelimit", "131072", CVAR_NOTFROMSERVER, "The maximum allowed file size of images loaded from a web-based url. 0 disables completely, while empty imposes no limit.");
extern cvar_t			scr_sshot_compression;
extern cvar_t gl_lerpimages;
extern cvar_t gl_picmip2d;
extern cvar_t gl_picmip;
extern cvar_t gl_picmip_world;
extern cvar_t gl_picmip_sprites;
extern cvar_t gl_picmip_other;
extern cvar_t r_shadow_bumpscale_basetexture;
extern cvar_t r_shadow_bumpscale_bumpmap;
extern cvar_t r_shadow_heightscale_basetexture;
extern cvar_t r_shadow_heightscale_bumpmap;


static bucket_t *imagetablebuckets[256];
static hashtable_t imagetable;
static image_t *imagelist;





#if defined(AVAIL_JPEGLIB) || defined(AVAIL_PNGLIB)
static void GenerateXMPData(char *blob, size_t blobsize, int width, int height, unsigned int metainfo)
{	//XMP is a general thing that applies to multiple formats - or at least png+jpeg.
	//we need this if we want to correctly flag the data as a 360 image.
	Q_snprintfz(blob, blobsize,
		"<x:xmpmeta xmlns:x='adobe:ns:meta/'>"
			"<rdf:RDF xmlns:rdf='http://www.w3.org/1999/02/22-rdf-syntax-ns#'>"
		);

	if (metainfo)
		Q_snprintfz(blob, blobsize,
					"<rdf:Description rdf:about='' xmlns:GPano=\"http://ns.google.com/photos/1.0/panorama/\">"
						"<GPano:ProjectionType>equirectangular</GPano:ProjectionType>"
						"<GPano:PosePitchDegrees>%f</GPano:PosePitchDegrees>"
						"<GPano:PoseHeadingDegrees>%f</GPano:PoseHeadingDegrees>"
						"<GPano:PoseRollDegrees>%f</GPano:PoseRollDegrees>"
						"<GPano:InitialViewHeadingDegrees>%f</GPano:InitialViewHeadingDegrees>"
						"<GPano:InitialViewPitchDegrees>%f</GPano:InitialViewPitchDegrees>"
						"<GPano:InitialViewRollDegrees>%f</GPano:InitialViewRollDegrees>"
						"<GPano:CroppedAreaLeftPixels>0</GPano:CroppedAreaLeftPixels>"
						"<GPano:CroppedAreaTopPixels>0</GPano:CroppedAreaTopPixels>"
						"<GPano:CroppedAreaImageWidthPixels>%i</GPano:CroppedAreaImageWidthPixels>"
						"<GPano:CroppedAreaImageHeightPixels>%i</GPano:CroppedAreaImageHeightPixels>"
						"<GPano:FullPanoWidthPixels>%i</GPano:FullPanoWidthPixels>"
						"<GPano:FullPanoHeightPixels>%i</GPano:FullPanoHeightPixels>"
					"</rdf:Description>",
			r_refdef.viewangles[0], r_refdef.viewangles[1], r_refdef.viewangles[2],
			r_refdef.viewangles[0], r_refdef.viewangles[1], r_refdef.viewangles[2],
			width, height, width, height);

	Q_snprintfz(blob+strlen(blob), blobsize-strlen(blob),
			"</rdf:RDF>"
		"</x:xmpmeta>"
		);
}
#endif
#endif

#ifndef _WIN32
#include <unistd.h>
#endif

typedef struct {	//cm = colourmap
	char	id_len;		//0
	char	cm_type;	//1
	char	version;	//2
		char pad1;
	short	cm_idx;		//3
	short	cm_len;		//5
	char	cm_size;	//7
		char pad2;
	short	originx;	//8 (ignored)
	short	originy;	//10 (ignored)
	short	width;		//12-13
	short	height;		//14-15
	qbyte	bpp;		//16
	qbyte	attribs;	//17
} tgaheader_t;

static char *ReadGreyTargaFile (qbyte *data, int flen, tgaheader_t *tgahead, int asgrey)	//preswapped header
{
	int				columns, rows;
	int				row, column;
	qbyte			*pixbuf, *pal;
	qboolean		flipped;

	qbyte *pixels = BZ_Malloc(tgahead->width * tgahead->height * (asgrey?1:4));

	if (tgahead->version!=1
		&& tgahead->version!=3)
	{
		Con_Printf("LoadGrayTGA: Only type 1 and 3 greyscale targa images are understood.\n");
		BZ_Free(pixels);
		return NULL;
	}

	if (tgahead->version==1 && tgahead->bpp != 8 &&
		tgahead->cm_size != 24 && tgahead->cm_len != 256)
	{
		Con_Printf("LoadGrayTGA: Strange palette type\n");
		BZ_Free(pixels);
		return NULL;
	}

	columns = tgahead->width;
	rows = tgahead->height;

	flipped = !((tgahead->attribs & 0x20) >> 5);
#ifndef NPFTE
	if (r_dodgytgafiles.value)
		flipped = true;
#endif

	if (tgahead->version == 1)
	{	//paletted data...
		pal = data;
		data += tgahead->cm_len*3;
		if (asgrey)
		{
			for(row=rows-1; row>=0; row--)
			{
				if (flipped)
					pixbuf = pixels + row*columns;
				else
					pixbuf = pixels + ((rows-1)-row)*columns;

				for(column=0; column<columns; column++)
					*pixbuf++= *data++;
			}
		}
		else
		{
			for(row=rows-1; row>=0; row--)
			{
				if (flipped)
					pixbuf = pixels + row*columns*4;
				else
					pixbuf = pixels + ((rows-1)-row)*columns*4;

				for(column=0; column<columns; column++)
				{
					*pixbuf++= pal[*data*3+2];
					*pixbuf++= pal[*data*3+1];
					*pixbuf++= pal[*data*3+0];
					*pixbuf++= 255;
					data++;
				}
			}
		}
		return pixels;
	}
	//version 3 now. pure greyscale

	if (asgrey)
	{
		for(row=rows-1; row>=0; row--)
		{
			if (flipped)
				pixbuf = pixels + row*columns;
			else
				pixbuf = pixels + ((rows-1)-row)*columns;

			pixbuf = pixels + row*columns;
			for(column=0; column<columns; column++)
				*pixbuf++= *data++;
		}
	}
	else
	{
		for(row=rows-1; row>=0; row--)
		{
			if (flipped)
				pixbuf = pixels + row*columns*4;
			else
				pixbuf = pixels + ((rows-1)-row)*columns*4;

			for(column=0; column<columns; column++)
			{
				*pixbuf++= *data;
				*pixbuf++= *data;
				*pixbuf++= *data;
				*pixbuf++= 255;
				data++;
			}
		}
	}

	return pixels;
}

#define MISSHORT(ptr) (*(ptr) | (*(ptr+1) << 8))
//remember to free it
//greyonly causes the function to fail if given anything but greyscale images
//this is for detecting heightmaps instead of normalmaps.
qbyte *ReadTargaFile(qbyte *buf, int length, int *width, int *height, uploadfmt_t *format, qboolean greyonly, uploadfmt_t forceformat)
{
	//tga files sadly lack a true magic header thing.
	unsigned char *data;

	qboolean flipped;

	tgaheader_t tgaheader;	//things are misaligned, so no pointer.

	if (length < 18 || buf[1] > 1 || (buf[16] != 8 && buf[16] != 16 && buf[16] != 24 && buf[16] != 32))
		return NULL;	//BUMMER!

	tgaheader.id_len = buf[0];
	tgaheader.cm_type = buf[1];
	tgaheader.version = buf[2];
	tgaheader.cm_idx = MISSHORT(buf+3);
	tgaheader.cm_len = MISSHORT(buf+5);
	tgaheader.cm_size = buf[7];
	tgaheader.originx = MISSHORT(buf+8);
	tgaheader.originy = MISSHORT(buf+10);
	tgaheader.width = MISSHORT(buf+12);
	tgaheader.height = MISSHORT(buf+14);
	tgaheader.bpp = buf[16];
	tgaheader.attribs = buf[17];

	switch(tgaheader.version)
	{
	case 0:	//No image data included.
		return NULL;	//not really valid for us. reject it after all
	case 1:	//Uncompressed, color-mapped images.
	case 2:	//Uncompressed, RGB images.
	case 3:	//Uncompressed, black and white images.
	case 9:	//Runlength encoded color-mapped images.
	case 10:	//Runlength encoded RGB images.
	case 11:	//Compressed, black and white images.
	case 32:	//Compressed color-mapped data, using Huffman, Delta, and runlength encoding.
	case 33:	//Compressed color-mapped data, using Huffman, Delta, and runlength encoding.  4-pass quadtree-type process.
		break;
	default:
		return NULL;
	}
	//validate the size to some sanity limit.
	if ((unsigned short)tgaheader.width > 16384 || (unsigned short)tgaheader.height > 16384)
		return NULL;


	flipped = !((tgaheader.attribs & 0x20) >> 5);
#ifndef NPFTE
	if (r_dodgytgafiles.value)
		flipped = true;
#endif

	data=buf+18;
	data += tgaheader.id_len;

	*width = tgaheader.width;
	*height = tgaheader.height;

	if (greyonly)	//grey only, load as 8 bit..
	{
		if (!(tgaheader.version == 1) && !(tgaheader.version == 3))
			return NULL;
	}
	if (tgaheader.version == 1 || tgaheader.version == 3)
	{
		if (forceformat==PTI_L8 || forceformat==PTI_RGBA8 || forceformat==PTI_RGBX8)
			*format = forceformat;
		else if (tgaheader.version == 3)
			*format = PTI_L8;
		else
			*format = PTI_RGBX8;
		return ReadGreyTargaFile(data, length, &tgaheader, *format==PTI_L8);
	}
	else if (tgaheader.version == 10 || tgaheader.version == 9 || tgaheader.version == 11)
	{
		//9:paletted
		//10:bgr(a)
		//11:greyscale
#undef getc
#define getc(x) *data++
		unsigned int row, rows=tgaheader.height, column, columns=tgaheader.width, packetHeader, packetSize, j;
		qbyte *pixbuf, *targa_rgba=BZ_Malloc(rows*columns*((forceformat==PTI_L8)?1:4)), *inrow;

		qbyte blue, red, green, alphabyte;

		byte_vec4_t palette[256];

		*format = PTI_RGBX8;
		if (tgaheader.version == 9)
		{	//RLE palette
			for (row = 0; row < 256; row++)
			{
				palette[row][0] = row;
				palette[row][1] = row;
				palette[row][2] = row;
				palette[row][3] = 255;
			}
			if (tgaheader.bpp != 8)	//FIXME: tgaheader.bpp can be 8, 15, or 16.
				return NULL;

			if (forceformat == PTI_L8 || forceformat == PTI_INVALID)
				*format = forceformat = PTI_L8;
			else
				*format = forceformat = PTI_LLLX8;


			if (tgaheader.cm_type)
			{
				qboolean grey = true;
				switch(tgaheader.cm_size)
				{
				case 24:
					for (row = 0; row < tgaheader.cm_len; row++)
					{
						if (data[0] != data[1] || data[0] != data[2])
							grey = false;
						palette[row][0] = *data++;
						palette[row][1] = *data++;
						palette[row][2] = *data++;
						palette[row][3] = 255;
					}
					if (grey && forceformat == PTI_INVALID)
						*format = PTI_L8;
					else if (forceformat != PTI_L8)
						*format = PTI_RGBA8;
					break;
				case 32:
					for (row = 0; row < tgaheader.cm_len; row++)
					{
						if (data[0] != data[1] || data[0] != data[2] || data[3] != 0xff)
							grey = false;
						palette[row][0] = *data++;
						palette[row][1] = *data++;
						palette[row][2] = *data++;
						palette[row][3] = *data++;
					}
					if (grey && forceformat == PTI_INVALID)
						*format = PTI_L8;
					else if (forceformat != PTI_L8)
						*format = PTI_RGBA8;
					break;
				}
			}
		}
		else if (tgaheader.version == 10)
		{	//RLE truecolour
			if (tgaheader.bpp == 8)
				return NULL;

			*format = (tgaheader.bpp==32)?PTI_RGBA8:PTI_RGBX8;
		}
		else if (tgaheader.version == 11)
		{	//RLE greyscale
			for (row = 0; row < 256; row++)
			{
				palette[row][0] = row;
				palette[row][1] = row;
				palette[row][2] = row;
				palette[row][3] = 255;
			}
			if (tgaheader.bpp != 8)
				return NULL;
		}

		for(row=rows; row-->0; )
		{
			if (flipped)
				pixbuf = targa_rgba + row*columns*((forceformat==PTI_L8)?1:4);
			else
				pixbuf = targa_rgba + ((rows-1)-row)*columns*((forceformat==PTI_L8)?1:4);
			for(column=0; column<columns; )
			{
				packetHeader=*data++;
				packetSize = 1 + (packetHeader & 0x7f);
				if (packetHeader & 0x80)
				{	// run-length packet
					switch (tgaheader.bpp)
					{
						case 8:	//we made sure this was version 11
								blue = palette[*data][0];
								green = palette[*data][1];
								red = palette[*data][2];
								alphabyte = palette[*data][3];
								data++;
								break;

						case 16:
								inrow = data;
								data+=2;
								red = ((inrow[1] & 0x7c)>>2) *8;					//red
								green =	(((inrow[1] & 0x03)<<3) + ((inrow[0] & 0xe0)>>5))*8;	//green
								blue = (inrow[0] & 0x1f)*8;					//blue
								alphabyte = (int)(inrow[1]&0x80)*2-1;			//alpha?
								break;
						case 24:
								blue = *data++;
								green = *data++;
								red = *data++;
								alphabyte = 255;
								break;
						case 32:
								blue = *data++;
								green = *data++;
								red = *data++;
								alphabyte = *data++;
								break;
						default:
								blue = 127;
								green = 127;
								red = 127;
								alphabyte = 127;
								break;
					}

					if (forceformat!=PTI_L8)	//keep colours
					{
						for(j=0;j<packetSize;j++)
						{
							*pixbuf++=red;
							*pixbuf++=green;
							*pixbuf++=blue;
							*pixbuf++=alphabyte;
							column++;
							if (column==columns)
							{ // run spans across rows
								column=0;
								if (row>0)
									row--;
								else
									goto breakOut;
								if (flipped)
									pixbuf = targa_rgba + row*columns*4;
								else
									pixbuf = targa_rgba + ((rows-1)-row)*columns*4;
							}
						}
					}
					else	//convert to greyscale
					{
						for(j=0;j<packetSize;j++)
						{
							*pixbuf++ = red*NTSC_RED + green*NTSC_GREEN + blue*NTSC_BLUE;
							column++;
							if (column==columns)
							{ // run spans across rows
								column=0;
								if (row>0)
									row--;
								else
									goto breakOut;
								if (flipped)
									pixbuf = targa_rgba + row*columns*1;
								else
									pixbuf = targa_rgba + ((rows-1)-row)*columns*1;
							}
						}
					}
				}
				else
				{                            // non run-length packet
					if (forceformat!=PTI_L8)	//keep colours
					{
						for(j=0;j<packetSize;j++)
						{
							switch (tgaheader.bpp)
							{
								case 8:
										blue = palette[*data][0];
										green = palette[*data][1];
										red = palette[*data][2];
										*pixbuf++ = red;
										*pixbuf++ = green;
										*pixbuf++ = blue;
										*pixbuf++ = palette[*data][3];
										data++;
										break;
								case 16:
										inrow = data;
										data+=2;
										red = ((inrow[1] & 0x7c)>>2) *8;					//red
										green =	(((inrow[1] & 0x03)<<3) + ((inrow[0] & 0xe0)>>5))*8;	//green
										blue = (inrow[0] & 0x1f)*8;					//blue
										alphabyte = (int)(inrow[1]&0x80)*2-1;			//alpha?

										*pixbuf++ = red;
										*pixbuf++ = green;
										*pixbuf++ = blue;
										*pixbuf++ = alphabyte;
										break;
								case 24:
										blue = *data++;
										green = *data++;
										red = *data++;
										*pixbuf++ = red;
										*pixbuf++ = green;
										*pixbuf++ = blue;
										*pixbuf++ = 255;
										break;
								case 32:
										blue = *data++;
										green = *data++;
										red = *data++;
										alphabyte = *data++;
										*pixbuf++ = red;
										*pixbuf++ = green;
										*pixbuf++ = blue;
										*pixbuf++ = alphabyte;
										break;
								default:
										blue = 127;
										green = 127;
										red = 127;
										alphabyte = 127;
										break;
							}
							column++;
							if (column==columns)
							{ // pixel packet run spans across rows
								column=0;
								if (row>0)
									row--;
								else
									goto breakOut;
								if (flipped)
									pixbuf = targa_rgba + row*columns*4;
								else
									pixbuf = targa_rgba + ((rows-1)-row)*columns*4;
							}
						}
					}
					else	//convert to grey
					{
						for(j=0;j<packetSize;j++)
						{
							switch (tgaheader.bpp)
							{
								case 8:
										blue = palette[*data][0];
										green = palette[*data][1];
										red = palette[*data][2];
										*pixbuf++ = (blue + green + red)/3;
										data++;
										break;
								case 16:
										inrow = data;
										data+=2;
										red = ((inrow[1] & 0x7c)>>2) *8;					//red
										green =	(((inrow[1] & 0x03)<<3) + ((inrow[0] & 0xe0)>>5))*8;	//green
										blue = (inrow[0] & 0x1f)*8;					//blue
										alphabyte = (int)(inrow[1]&0x80)*2-1;			//alpha?

										*pixbuf++ = red*NTSC_RED + green*NTSC_GREEN + blue*NTSC_BLUE;
										break;
								case 24:
										blue = *data++;
										green = *data++;
										red = *data++;
										*pixbuf++ = red*NTSC_RED + green*NTSC_GREEN + blue*NTSC_BLUE;
										break;
								case 32:
										blue = *data++;
										green = *data++;
										red = *data++;
										alphabyte = *data++;
										*pixbuf++ = red*NTSC_RED + green*NTSC_GREEN + blue*NTSC_BLUE;
										break;
								default:
										blue = 127;
										green = 127;
										red = 127;
										alphabyte = 127;
										break;
							}
							column++;
							if (column==columns)
							{ // pixel packet run spans across rows
								column=0;
								if (row>0)
									row--;
								else
									goto breakOut;
								if (flipped)
									pixbuf = targa_rgba + row*columns*1;
								else
									pixbuf = targa_rgba + ((rows-1)-row)*columns*1;
							}
						}
					}
				}
			}
		}
		breakOut:;

		return targa_rgba;
	}
	else if (tgaheader.version == 2)
	{	//packed format
		qbyte *initbuf=BZ_Malloc(tgaheader.height*tgaheader.width* ((forceformat==PTI_L8)?1:4));
		qbyte *inrow, *outrow;
		int x, y, mul;
		qbyte blue, red, green;

		if (tgaheader.bpp == 8)
			return NULL;

		mul = tgaheader.bpp/8;
//flip +convert to 32 bit
		if (forceformat==PTI_L8)
		{
			*format = forceformat;
			outrow = &initbuf[(int)(0)*tgaheader.width];
		}
		else
		{
			outrow = &initbuf[(int)(0)*tgaheader.width*mul];
			*format = (mul==4)?PTI_RGBA8:PTI_RGBX8;
		}
		for (y = 0; y < tgaheader.height; y+=1)
		{
			if (flipped)
				inrow = &data[(int)(tgaheader.height-y-1)*tgaheader.width*mul];
			else
				inrow = &data[(int)(y)*tgaheader.width*mul];

			if (forceformat!=PTI_L8)
			{
				switch(mul)
				{
				case 2:
					for (x = 0; x < tgaheader.width; x+=1)
					{
						*outrow++ = ((inrow[1] & 0x7c)>>2) *8;					//red
						*outrow++ = (((inrow[1] & 0x03)<<3) + ((inrow[0] & 0xe0)>>5))*8;	//green
						*outrow++ = (inrow[0] & 0x1f)*8;					//blue
						*outrow++ = (int)(inrow[1]&0x80)*2-1;			//alpha?
						inrow+=2;
					}
					break;
				case 3:
					for (x = 0; x < tgaheader.width; x+=1)
					{
						*outrow++ = inrow[2];
						*outrow++ = inrow[1];
						*outrow++ = inrow[0];
						*outrow++ = 255;
						inrow+=3;
					}
					break;
				case 4:
					for (x = 0; x < tgaheader.width; x+=1)
					{
						*outrow++ = inrow[2];
						*outrow++ = inrow[1];
						*outrow++ = inrow[0];
						*outrow++ = inrow[3];
						inrow+=4;
					}
					break;
				}
			}
			else
			{
				switch(mul)
				{
				case 2:
					for (x = 0; x < tgaheader.width; x+=1)
					{
						red = ((inrow[1] & 0x7c)>>2) *8;					//red
						green = (((inrow[1] & 0x03)<<3) + ((inrow[0] & 0xe0)>>5))*8;	//green
						blue = (inrow[0] & 0x1f)*8;					//blue
//						alphabyte = (int)(inrow[1]&0x80)*2-1;			//alpha?

						*outrow++ = red*NTSC_RED + green*NTSC_GREEN + blue*NTSC_BLUE;
						inrow+=2;
					}
					break;
				case 3:
					for (x = 0; x < tgaheader.width; x+=1)
					{
						red = inrow[2];
						green = inrow[1];
						blue = inrow[0];
						*outrow++ = red*NTSC_RED + green*NTSC_GREEN + blue*NTSC_BLUE;
						inrow+=3;
					}
					break;
				case 4:
					for (x = 0; x < tgaheader.width; x+=1)
					{
						red = inrow[2];
						green = inrow[1];
						blue = inrow[0];
						*outrow++ = red*NTSC_RED + green*NTSC_GREEN + blue*NTSC_BLUE;
						inrow+=4;
					}
					break;
				}
			}
		}

		if (forceformat!=PTI_L8)
		{
			for (x = 0; x < tgaheader.width*tgaheader.height*4; x+=4)
			{
				if (initbuf[x+0] != initbuf[x+1] || initbuf[x+0] != initbuf[x+2] || initbuf[x+3] != 0xff)
					break;
			}
			if (x == tgaheader.width*tgaheader.height*4)
			{	//no alpha
				if (forceformat==PTI_INVALID)
					*format = PTI_LLLX8;
			}
			else
			{
				for (; x < tgaheader.width*tgaheader.height*4; x+=4)
				{
					if (initbuf[x+0] != initbuf[x+1] || initbuf[x+0] != initbuf[x+2])
						break;
				}
				if (x == tgaheader.width*tgaheader.height*4)
				{	//okay, there's some alpha data in there.
					if (forceformat==PTI_INVALID)
						*format = PTI_LLLA8;
				}
			}
		}
		return initbuf;
	}
	else
		Con_Printf("TGA: Unsupported version\n");
	return NULL;
}

qboolean WriteTGA(char *filename, enum fs_relative fsroot, const qbyte *fte_restrict rgb_buffer, qintptr_t bytestride, int width, int height, enum uploadfmt fmt)
{
	qboolean success = false;
	size_t c, i;
	vfsfile_t *vfs;
	if (fmt != TF_BGRA32 && fmt != TF_RGB24 && fmt != TF_RGBA32 && fmt != TF_BGR24 && fmt != TF_RGBX32 && fmt != TF_BGRX32)
		return false;
	FS_CreatePath(filename, fsroot);
	vfs = FS_OpenVFS(filename, "wb", fsroot);
	if (vfs)
	{
		int ipx,opx;
		qboolean rgb;
		unsigned char header[18];
		memset (header, 0, 18);

		if (fmt == TF_BGRA32 || fmt == TF_RGBA32)
		{
			rgb = fmt==TF_RGBA32;
			ipx = 4;
			opx = 4;
		}
		else if (fmt == TF_RGBX32 || fmt == TF_BGRX32)
		{
			rgb = fmt==TF_RGBX32;
			ipx = 4;
			opx = 3;
		}
		else
		{
			rgb = fmt==TF_RGB24;
			ipx = 3;
			opx = 3;
		}

		header[2] = 2;			// uncompressed type
		header[12] = width&255;
		header[13] = width>>8;
		header[14] = height&255;
		header[15] = height>>8;
		header[16] = opx*8;		// pixel size
		header[17] = 0x00;		// flags

		if (bytestride < 0)
		{	//if we're upside down, lets just use an upside down tga.
			rgb_buffer += bytestride*(height-1);
			bytestride = -bytestride;
			//now we can just do everything without worrying about rows
		}
		else	//our data is top-down, set up the header to also be top-down.
			header[17] = 0x20;

		if (ipx == opx && !rgb)
		{	//can just directly write it
			//bgr24, bgra24
			c = (size_t)width*height*opx;

			VFS_WRITE(vfs, header, sizeof(header));
			VFS_WRITE(vfs, rgb_buffer, c);
		}
		else
		{
			qbyte *fte_restrict rgb_out = malloc((size_t)width*opx*height);

			//no need to swap alpha, and if we're just swapping alpha will be fine in-place.
			if (rgb)
			{	//rgb24, rgbx32, rgba32
				// compact, and swap
				c = (size_t)width*height;
				for (i=0 ; i<c ; i++)
				{
					rgb_out[i*opx+2] = rgb_buffer[i*ipx+0];
					rgb_out[i*opx+1] = rgb_buffer[i*ipx+1];
					rgb_out[i*opx+0] = rgb_buffer[i*ipx+2];
				}
			}
			else
			{	//(bgr24), bgrx32, (bgra32)
				// compact
				c = (size_t)width*height;
				for (i=0 ; i<c ; i++)
				{
					rgb_out[i*opx+0] = rgb_buffer[i*ipx+0];
					rgb_out[i*opx+1] = rgb_buffer[i*ipx+1];
					rgb_out[i*opx+2] = rgb_buffer[i*ipx+2];
				}
			}
			c *= opx;

			VFS_WRITE(vfs, header, sizeof(header));
			VFS_WRITE(vfs, rgb_out, c);
			free(rgb_out);
		}

		success = VFS_CLOSE(vfs);
	}
	return success;
}

#ifdef AVAIL_PNGLIB
	#ifndef AVAIL_ZLIB
		#error PNGLIB requires ZLIB
	#endif

	#undef channels

	#ifndef PNG_SUCKS_WITH_SETJMP
		#if defined(MINGW)
			#include "./mingw-libs/png.h"
		#elif defined(_WIN32)
			#include "png.h"
		#else
			#include <png.h>
		#endif
	#endif

	#ifdef DYNAMIC_LIBPNG
		#define PSTATIC(n)
		static dllhandle_t *libpng_handle;
		#define LIBPNG_LOADED() (libpng_handle != NULL)
	#else
		#define LIBPNG_LOADED() 1
		#define PSTATIC(n) = &n
		#ifdef _MSC_VER
			#ifdef _WIN64
				#pragma comment(lib, MSVCLIBSPATH "libpng64.lib")
			#else
				#pragma comment(lib, MSVCLIBSPATH "libpng.lib")
			#endif
		#endif
	#endif

#ifndef PNG_NORETURN
#define PNG_NORETURN
#endif
#ifndef PNG_ALLOCATED
#define PNG_ALLOCATED
#endif

#if PNG_LIBPNG_VER < 10500
	#define png_const_infop png_infop
	#define png_const_structp png_structp
	#define png_const_bytep png_bytep
	#define png_const_unknown_chunkp png_unknown_chunkp
	#define png_const_textp png_textp
#endif
#if PNG_LIBPNG_VER < 10600
	#define png_inforp png_infop
	#define png_const_inforp png_const_infop
	#define png_structrp png_structp
	#define png_const_structrp png_const_structp
#endif

static void (PNGAPI *qpng_error) PNGARG((png_const_structrp png_ptr, png_const_charp error_message)) PSTATIC(png_error);
static void (PNGAPI *qpng_read_end) PNGARG((png_structp png_ptr, png_infop info_ptr)) PSTATIC(png_read_end);
static void (PNGAPI *qpng_read_image) PNGARG((png_structp png_ptr, png_bytepp image)) PSTATIC(png_read_image);
static png_byte (PNGAPI *qpng_get_bit_depth) PNGARG((png_const_structp png_ptr, png_const_inforp info_ptr)) PSTATIC(png_get_bit_depth);
static png_byte (PNGAPI *qpng_get_channels) PNGARG((png_const_structp png_ptr, png_const_inforp info_ptr)) PSTATIC(png_get_channels);
#if PNG_LIBPNG_VER < 10400
	static png_uint_32 (PNGAPI *qpng_get_rowbytes) PNGARG((png_const_structp png_ptr, png_const_inforp info_ptr)) PSTATIC(png_get_rowbytes);
#else
	static png_size_t (PNGAPI *qpng_get_rowbytes) PNGARG((png_const_structp png_ptr, png_const_inforp info_ptr)) PSTATIC(png_get_rowbytes);
#endif
static void (PNGAPI *qpng_read_update_info) PNGARG((png_structp png_ptr, png_infop info_ptr)) PSTATIC(png_read_update_info);
static void (PNGAPI *qpng_set_strip_16) PNGARG((png_structp png_ptr)) PSTATIC(png_set_strip_16);
static void (PNGAPI *qpng_set_swap) PNGARG((png_structp png_ptr)) PSTATIC(png_set_swap);
static void (PNGAPI *qpng_set_expand) PNGARG((png_structp png_ptr)) PSTATIC(png_set_expand);
static void (PNGAPI *qpng_set_gray_to_rgb) PNGARG((png_structp png_ptr)) PSTATIC(png_set_gray_to_rgb);
static void (PNGAPI *qpng_set_tRNS_to_alpha) PNGARG((png_structp png_ptr)) PSTATIC(png_set_tRNS_to_alpha);
static png_uint_32 (PNGAPI *qpng_get_valid) PNGARG((png_const_structp png_ptr, png_const_infop info_ptr, png_uint_32 flag)) PSTATIC(png_get_valid);
#if PNG_LIBPNG_VER >= 10400
static void (PNGAPI *qpng_set_expand_gray_1_2_4_to_8) PNGARG((png_structp png_ptr)) PSTATIC(png_set_expand_gray_1_2_4_to_8);
#else
static void (PNGAPI *qpng_set_gray_1_2_4_to_8) PNGARG((png_structp png_ptr)) PSTATIC(png_set_gray_1_2_4_to_8);
#endif
static void (PNGAPI *qpng_set_bgr) PNGARG((png_structp png_ptr)) PSTATIC(png_set_bgr);
static void (PNGAPI *qpng_set_filler) PNGARG((png_structp png_ptr, png_uint_32 filler, int flags)) PSTATIC(png_set_filler);
static void (PNGAPI *qpng_set_palette_to_rgb) PNGARG((png_structp png_ptr)) PSTATIC(png_set_palette_to_rgb);
static png_uint_32 (PNGAPI *qpng_get_IHDR) PNGARG((png_const_structrp png_ptr, png_const_inforp info_ptr, png_uint_32 *width, png_uint_32 *height,
			int *bit_depth, int *color_type, int *interlace_method, int *compression_method, int *filter_method)) PSTATIC(png_get_IHDR);
static void (PNGAPI *qpng_read_info) PNGARG((png_structp png_ptr, png_infop info_ptr)) PSTATIC(png_read_info);
static void (PNGAPI *qpng_set_sig_bytes) PNGARG((png_structp png_ptr, int num_bytes)) PSTATIC(png_set_sig_bytes);
static void (PNGAPI *qpng_set_read_fn) PNGARG((png_structp png_ptr, png_voidp io_ptr, png_rw_ptr read_data_fn)) PSTATIC(png_set_read_fn);
static void (PNGAPI *qpng_destroy_read_struct) PNGARG((png_structpp png_ptr_ptr, png_infopp info_ptr_ptr, png_infopp end_info_ptr_ptr)) PSTATIC(png_destroy_read_struct);
static png_infop (PNGAPI *qpng_create_info_struct) PNGARG((png_const_structrp png_ptr)) PSTATIC(png_create_info_struct);
static png_structp (PNGAPI *qpng_create_read_struct) PNGARG((png_const_charp user_png_ver, png_voidp error_ptr, png_error_ptr error_fn, png_error_ptr warn_fn)) PSTATIC(png_create_read_struct);
static int (PNGAPI *qpng_sig_cmp) PNGARG((png_const_bytep sig, png_size_t start, png_size_t num_to_check)) PSTATIC(png_sig_cmp);

static void (PNGAPI *qpng_write_end) PNGARG((png_structrp png_ptr, png_inforp info_ptr)) PSTATIC(png_write_end);
static void (PNGAPI *qpng_write_image) PNGARG((png_structrp png_ptr, png_bytepp image)) PSTATIC(png_write_image);
static void (PNGAPI *qpng_write_info) PNGARG((png_structrp png_ptr, png_const_inforp info_ptr)) PSTATIC(png_write_info);
#ifdef PNG_TEXT_SUPPORTED
static void (PNGAPI *qpng_set_text) PNGARG((png_const_structrp png_ptr, png_infop info_ptr, png_const_textp text_ptr, int num_text)) PSTATIC(png_set_text);
#endif
static void (PNGAPI *qpng_set_IHDR) PNGARG((png_const_structrp png_ptr, png_infop info_ptr, png_uint_32 width, png_uint_32 height,
			int bit_depth, int color_type, int interlace_method, int compression_method, int filter_method)) PSTATIC(png_set_IHDR);
static void (PNGAPI *qpng_set_compression_level) PNGARG((png_structrp png_ptr, int level)) PSTATIC(png_set_compression_level);
static void (PNGAPI *qpng_init_io) PNGARG((png_structp png_ptr, png_FILE_p fp)) PSTATIC(png_init_io);
static png_voidp (PNGAPI *qpng_get_io_ptr) PNGARG((png_const_structrp png_ptr)) PSTATIC(png_get_io_ptr);
static void (PNGAPI *qpng_destroy_write_struct) PNGARG((png_structpp png_ptr_ptr, png_infopp info_ptr_ptr)) PSTATIC(png_destroy_write_struct);
static png_structp (PNGAPI *qpng_create_write_struct) PNGARG((png_const_charp user_png_ver, png_voidp error_ptr, png_error_ptr error_fn, png_error_ptr warn_fn)) PSTATIC(png_create_write_struct);
static void (PNGAPI *qpng_set_unknown_chunks) PNGARG((png_const_structrp png_ptr, png_inforp info_ptr, png_const_unknown_chunkp unknowns, int num_unknowns)) PSTATIC(png_set_unknown_chunks);

static png_voidp (PNGAPI *qpng_get_error_ptr) PNGARG((png_const_structrp png_ptr)) PSTATIC(png_get_error_ptr);

qboolean LibPNG_Init(void)
{
#ifdef DYNAMIC_LIBPNG
	static dllfunction_t pngfuncs[] =
	{
		{(void **) &qpng_error,							"png_error"},
		{(void **) &qpng_read_end,						"png_read_end"},
		{(void **) &qpng_read_image,					"png_read_image"},
		{(void **) &qpng_get_bit_depth,					"png_get_bit_depth"},
		{(void **) &qpng_get_channels,					"png_get_channels"},
		{(void **) &qpng_get_rowbytes,					"png_get_rowbytes"},
		{(void **) &qpng_read_update_info,				"png_read_update_info"},
		{(void **) &qpng_set_strip_16,					"png_set_strip_16"},
		{(void **) &qpng_set_swap,						"png_set_swap"},
		{(void **) &qpng_set_expand,					"png_set_expand"},
		{(void **) &qpng_set_gray_to_rgb,				"png_set_gray_to_rgb"},
		{(void **) &qpng_set_tRNS_to_alpha,				"png_set_tRNS_to_alpha"},
		{(void **) &qpng_get_valid,						"png_get_valid"},
#if PNG_LIBPNG_VER > 10400
		{(void **) &qpng_set_expand_gray_1_2_4_to_8,	"png_set_expand_gray_1_2_4_to_8"},
#else
		{(void **) &qpng_set_gray_1_2_4_to_8,	"png_set_gray_1_2_4_to_8"},
#endif
		{(void **) &qpng_set_bgr,						"png_set_bgr"},
		{(void **) &qpng_set_filler,					"png_set_filler"},
		{(void **) &qpng_set_palette_to_rgb,			"png_set_palette_to_rgb"},
		{(void **) &qpng_get_IHDR,						"png_get_IHDR"},
		{(void **) &qpng_read_info,						"png_read_info"},
		{(void **) &qpng_set_sig_bytes,					"png_set_sig_bytes"},
		{(void **) &qpng_set_read_fn,					"png_set_read_fn"},
		{(void **) &qpng_destroy_read_struct,			"png_destroy_read_struct"},
		{(void **) &qpng_create_info_struct,			"png_create_info_struct"},
		{(void **) &qpng_create_read_struct,			"png_create_read_struct"},
		{(void **) &qpng_sig_cmp,						"png_sig_cmp"},

#ifdef PNG_TEXT_SUPPORTED
		{(void **) &qpng_set_text,						"png_set_text"},
#endif
		{(void **) &qpng_write_end,						"png_write_end"},
		{(void **) &qpng_write_image,					"png_write_image"},
		{(void **) &qpng_write_info,					"png_write_info"},
		{(void **) &qpng_set_IHDR,						"png_set_IHDR"},
		{(void **) &qpng_set_compression_level,			"png_set_compression_level"},
		{(void **) &qpng_init_io,						"png_init_io"},
		{(void **) &qpng_get_io_ptr,					"png_get_io_ptr"},
		{(void **) &qpng_destroy_write_struct,			"png_destroy_write_struct"},
		{(void **) &qpng_create_write_struct,			"png_create_write_struct"},
		{(void **) &qpng_set_unknown_chunks,			"png_set_unknown_chunks"},

		{(void **) &qpng_get_error_ptr,					"png_get_error_ptr"},
		{NULL, NULL}
	};
	static qboolean tried;
	if (!tried)
	{
		tried = true;

		if (!LIBPNG_LOADED())
		{
			char *libnames[] =
			{
			#ifdef _WIN32
				va("libpng%i", PNG_LIBPNG_VER_DLLNUM)
			#else
				//linux...
				//lsb uses 'libpng12.so' specifically, so make sure that works.
				"libpng" STRINGIFY(PNG_LIBPNG_VER_MAJOR) STRINGIFY(PNG_LIBPNG_VER_MINOR) ".so." STRINGIFY(PNG_LIBPNG_VER_SONUM),
				"libpng" STRINGIFY(PNG_LIBPNG_VER_MAJOR) STRINGIFY(PNG_LIBPNG_VER_MINOR) ".so",
				"libpng.so." STRINGIFY(PNG_LIBPNG_VER_SONUM)
				"libpng.so",
			#endif
			};
			size_t i;
			for (i = 0; i < countof(libnames); i++)
			{
				if (libnames[i])
				{
					libpng_handle = Sys_LoadLibrary(libnames[i], pngfuncs);
					if (libpng_handle)
						break;
				}
			}
			if (!libpng_handle)
				Con_Printf("Unable to load %s\n", libnames[0]);
		}

//		if (!LIBPNG_LOADED())
//			libpng_handle = Sys_LoadLibrary("libpng", pngfuncs);
	}
#endif
	return LIBPNG_LOADED();
}

typedef struct {
	char *data;
	int readposition;
	int filelen;
} pngreadinfo_t;

static void VARGS readpngdata(png_structp png_ptr,png_bytep data,png_size_t len)
{
	pngreadinfo_t *ri = (pngreadinfo_t*)qpng_get_io_ptr(png_ptr);
	if (ri->readposition+len > ri->filelen)
	{
		qpng_error(png_ptr, "unexpected eof");
		return;
	}
	memcpy(data, &ri->data[ri->readposition], len);
	ri->readposition+=len;
}

struct pngerr
{
	const char *fname;
	jmp_buf jbuf;
};
static void VARGS png_onerror(png_structp png_ptr, png_const_charp error_msg)
{
	struct pngerr *err = qpng_get_error_ptr(png_ptr);
	Con_Printf("libpng %s: %s\n", err->fname, error_msg);
	longjmp(err->jbuf, 1);
	abort();
}

static void VARGS png_onwarning(png_structp png_ptr, png_const_charp warning_msg)
{
	struct pngerr *err = qpng_get_error_ptr(png_ptr);
#ifndef NPFTE
	Con_DPrintf("libpng %s: %s\n", err->fname, warning_msg);
#endif
}

qbyte *ReadPNGFile(const char *fname, qbyte *buf, int length, int *width, int *height, uploadfmt_t *format)
{
	qbyte header[8], **rowpointers = NULL, *data = NULL;
	png_structp png;
	png_infop pnginfo;
	int y, bitdepth, colortype, interlace, compression, filter, channels;
	unsigned long rowbytes;
	pngreadinfo_t ri;
	png_uint_32 pngwidth, pngheight;
	struct pngerr errctx;

	if (!LibPNG_Init())
		return NULL;

	memcpy(header, buf, 8);

	errctx.fname = fname;
	if (setjmp(errctx.jbuf))
	{
error:
		if (data)
			BZ_Free(data);
		if (rowpointers)
			BZ_Free(rowpointers);
		qpng_destroy_read_struct(&png, &pnginfo, NULL);
		return NULL;
	}

	if (qpng_sig_cmp(header, 0, 8))
	{
		return NULL;
	}

	if (!(png = qpng_create_read_struct(PNG_LIBPNG_VER_STRING, &errctx, png_onerror, png_onwarning)))
	{
		return NULL;
	}

	if (!(pnginfo = qpng_create_info_struct(png)))
	{
		qpng_destroy_read_struct(&png, &pnginfo, NULL);
		return NULL;
	}

	ri.data=buf;
	ri.readposition=8;
	ri.filelen=length;
	qpng_set_read_fn(png, &ri, readpngdata);

	qpng_set_sig_bytes(png, 8);
	qpng_read_info(png, pnginfo);
	qpng_get_IHDR(png, pnginfo, &pngwidth, &pngheight, &bitdepth, &colortype, &interlace, &compression, &filter);

	*width = pngwidth;
	*height = pngheight;

	if (colortype == PNG_COLOR_TYPE_PALETTE)
	{
		qpng_set_palette_to_rgb(png);
		qpng_set_filler(png, ~0u, PNG_FILLER_AFTER);
	}

	if (colortype == PNG_COLOR_TYPE_GRAY && bitdepth < 8)
	{
		#if PNG_LIBPNG_VER > 10400
			qpng_set_expand_gray_1_2_4_to_8(png);
		#else
			qpng_set_gray_1_2_4_to_8(png);
		#endif
	}

	if (qpng_get_valid( png, pnginfo, PNG_INFO_tRNS))
		qpng_set_tRNS_to_alpha(png);

	if (bitdepth >= 8 && colortype == PNG_COLOR_TYPE_RGB)
		qpng_set_filler(png, ~0u, PNG_FILLER_AFTER);

	if (colortype == PNG_COLOR_TYPE_GRAY || colortype == PNG_COLOR_TYPE_GRAY_ALPHA)
	{
		qpng_set_gray_to_rgb( png );
		qpng_set_filler(png, ~0u, PNG_FILLER_AFTER);
	}

	if (bitdepth < 8)
		qpng_set_expand (png);
	else if (bitdepth == 16 && !format)
		qpng_set_strip_16(png);
	else if (bitdepth == 16)
        qpng_set_swap(png);

	qpng_read_update_info(png, pnginfo);
	rowbytes = qpng_get_rowbytes(png, pnginfo);
	channels = qpng_get_channels(png, pnginfo);
	bitdepth = qpng_get_bit_depth(png, pnginfo);

	if (bitdepth == 8 && channels == 4)
	{
		if (format)
			*format = PTI_RGBA8;
	}
	else if (bitdepth == 16 && channels == 4)
		*format = PTI_RGBA16;
	else
	{
		Con_Printf ("Bad PNG color depth and/or bpp (%s)\n", fname);
		qpng_destroy_read_struct(&png, &pnginfo, NULL);
		return NULL;
	}

	data = BZF_Malloc(*height * rowbytes);
	rowpointers = BZF_Malloc(*height * sizeof(*rowpointers));

	if (!data || !rowpointers)
		goto error;

	for (y = 0; y < *height; y++)
		rowpointers[y] = data + y * rowbytes;

	qpng_read_image(png, rowpointers);
	qpng_read_end(png, NULL);

	qpng_destroy_read_struct(&png, &pnginfo, NULL);
	BZ_Free(rowpointers);
	return data;
}



#ifndef NPFTE
static int Image_WritePNG (char *filename, enum fs_relative fsroot, int compression, void **buffers, int numbuffers, qintptr_t bufferstride, int width, int height, enum uploadfmt fmt, qboolean writemetadata)
{
	char name[MAX_OSPATH];
	int i;
	FILE *fp;
	png_structp png_ptr;
	png_infop info_ptr;
	png_byte **row_pointers;
	struct pngerr errctx;
	int pxsize;
	int outwidth = width;

	qbyte stereochunk = 0;	//cross-eyed
	png_unknown_chunk unknowns = {"sTER", &stereochunk, sizeof(stereochunk), PNG_HAVE_PLTE};

	if (!FS_NativePath(filename, fsroot, name, sizeof(name)))
		return false;

	if (numbuffers == 2)
	{
		outwidth = width;
		if (outwidth & 7)	//standard stereo images must be padded to 8 pixels width padding between
			outwidth += 8-(outwidth & 7);
		outwidth += width;
	}
	else	//arrange them all horizontally
		outwidth = width * numbuffers;

	if (!LibPNG_Init())
		return false;

	if (!(fp = fopen (name, "wb")))
	{
		FS_CreatePath (filename, FS_GAMEONLY);
		if (!(fp = fopen (name, "wb")))
			return false;
	}

	errctx.fname = filename;
	if (setjmp(errctx.jbuf))
	{
err:
		qpng_destroy_write_struct(&png_ptr, &info_ptr);
		fclose(fp);
		return false;
	}

	if (!(png_ptr = qpng_create_write_struct(PNG_LIBPNG_VER_STRING, &errctx, png_onerror, png_onwarning)))
	{
		fclose(fp);
		return false;
	}

	if (!(info_ptr = qpng_create_info_struct(png_ptr)))
	{
		qpng_destroy_write_struct(&png_ptr, (png_infopp) NULL);
		fclose(fp);
		return false;
	}

	qpng_init_io(png_ptr, fp);
	compression = bound(0, compression, 100);

// had to add these when I migrated from libpng 1.4.x to 1.5.x
#ifndef Z_NO_COMPRESSION
#define Z_NO_COMPRESSION			0
#endif
#ifndef Z_BEST_COMPRESSION
#define Z_BEST_COMPRESSION			9
#endif
	qpng_set_compression_level(png_ptr, Z_NO_COMPRESSION + (compression*(Z_BEST_COMPRESSION-Z_NO_COMPRESSION))/100);

	if (fmt == TF_BGR24 || fmt == TF_BGRA32 || fmt == TF_BGRX32)
		qpng_set_bgr(png_ptr);
	if (fmt == TF_RGBA32 || fmt == TF_BGRA32 || fmt == PTI_LLLA8)
	{
		pxsize = 4;
		qpng_set_IHDR(png_ptr, info_ptr, outwidth, height, 8, PNG_COLOR_TYPE_RGBA, PNG_INTERLACE_NONE, PNG_COMPRESSION_TYPE_DEFAULT, PNG_FILTER_TYPE_DEFAULT);
	}
	else if (fmt == TF_RGBX32 || fmt == TF_BGRX32 || fmt == PTI_LLLX8)
	{
		pxsize = 4;
		qpng_set_IHDR(png_ptr, info_ptr, outwidth, height, 8, PNG_COLOR_TYPE_RGB, PNG_INTERLACE_NONE, PNG_COMPRESSION_TYPE_DEFAULT, PNG_FILTER_TYPE_DEFAULT);
	}
	else
	{
		pxsize = 3;
		qpng_set_IHDR(png_ptr, info_ptr, outwidth, height, 8, PNG_COLOR_TYPE_RGB, PNG_INTERLACE_NONE, PNG_COMPRESSION_TYPE_DEFAULT, PNG_FILTER_TYPE_DEFAULT);
	}

#ifdef PNG_TEXT_SUPPORTED
	if (writemetadata)
	{
		char blob[8192];
		png_text pngtext = {PNG_ITXT_COMPRESSION_NONE, "XML:com.adobe.xmp"};
		pngtext.text = blob;
		GenerateXMPData(blob, sizeof(blob), width, height, writemetadata);
		pngtext.itxt_length = strlen(pngtext.text);
		qpng_set_text(png_ptr, info_ptr, &pngtext, 1);
	}
#endif

	if (numbuffers == 2)	//flag it as a standard stereographic image
		qpng_set_unknown_chunks(png_ptr, info_ptr, &unknowns, 1);

	qpng_write_info(png_ptr, info_ptr);
	if (fmt == TF_RGBX32 || fmt == TF_BGRX32)
		qpng_set_filler(png_ptr, 0, PNG_FILLER_AFTER);

	if (numbuffers == 2)
	{	//standard stereographic png image.
		qbyte *pixels, *left, *right;
		//we need to pack the data into a single image for libpng to use
		row_pointers = Z_Malloc (sizeof(png_byte *) * height + outwidth*height*pxsize);	//must be zeroed, because I'm too lazy to specially deal with padding.
		if (!row_pointers)
			goto err;
		pixels = (qbyte*)row_pointers + height;
		//png requires right then left, which is a bit weird.
		//they're meant to be viewable by going cross-eyed (if needed)
		right = pixels;
		left = right + (outwidth-width)*pxsize;

		for (i = 0; i < height; i++)
		{
			if ((qbyte*)buffers[1])
				memcpy(right + i*outwidth*pxsize, (qbyte*)buffers[1] + i*bufferstride, pxsize * width);
			if ((qbyte*)buffers[0])
				memcpy(left + i*outwidth*pxsize, (qbyte*)buffers[0] + i*bufferstride, pxsize * width);
			row_pointers[i] = pixels + i * outwidth * pxsize;
		}
	}
	else if (numbuffers == 1)
	{
		row_pointers = BZ_Malloc (sizeof(png_byte *) * height);
		if (!row_pointers)
			goto err;
		for (i = 0; i < height; i++)
			row_pointers[i] = (qbyte*)buffers[0] + i * bufferstride;
	}
	else
	{	//pack all images horizontally, because preventing people from doing the whole cross-eyed thing is cool, or something.
		qbyte *pixels;
		int j;
		//we need to pack the data into a single image for libpng to use
		row_pointers = BZ_Malloc (sizeof(png_byte *) * height + outwidth*height*pxsize);
		if (!row_pointers)
			goto err;
		pixels = (qbyte*)row_pointers + height;
		for (i = 0; i < height; i++)
		{
			for (j = 0; j < numbuffers; j++)
			{
				if (buffers[j])
					memcpy(pixels+(width*j + i*outwidth)*pxsize, (qbyte*)buffers[j] + i*bufferstride, pxsize * width);
			}
			row_pointers[i] = pixels + i * outwidth * pxsize;
		}
	}
	qpng_write_image(png_ptr, row_pointers);
	qpng_write_end(png_ptr, info_ptr);
	BZ_Free(row_pointers);
	qpng_destroy_write_struct(&png_ptr, &info_ptr);
	if (0==fclose(fp))
		return true;
	Con_Printf("File error writing %s\n", filename);
	return false;
}
#endif


#endif

#ifdef AVAIL_JPEGLIB
#define XMD_H	//fix for mingw

#if defined(MINGW)
	#define JPEG_API VARGS
	#include "./mingw-libs/jpeglib.h"
	#include "./mingw-libs/jerror.h"
#elif defined(_WIN32)
	#define JPEG_API VARGS
	#include "jpeglib.h"
	#include "jerror.h"
#else
//	#include <jinclude.h>
	#include <jpeglib.h>
	#include <jerror.h>
#endif

#ifdef DYNAMIC_LIBJPEG
	#define JSTATIC(n)
	static dllhandle_t *libjpeg_handle;
	#define LIBJPEG_LOADED() (libjpeg_handle != NULL)
#else
	#ifdef _MSC_VER
		#ifdef _WIN64
			#pragma comment(lib, MSVCLIBSPATH "libjpeg64.lib")
		#else
			#pragma comment(lib, MSVCLIBSPATH "jpeg.lib")
		#endif
	#endif
	#define JSTATIC(n) = &n
	#define LIBJPEG_LOADED() (1)
#endif

#ifndef JPEG_FALSE
#define JPEG_boolean boolean
#endif

#define qjpeg_create_compress(cinfo) \
    qjpeg_CreateCompress((cinfo), JPEG_LIB_VERSION, \
			(size_t) sizeof(struct jpeg_compress_struct))
#define qjpeg_create_decompress(cinfo) \
    qjpeg_CreateDecompress((cinfo), JPEG_LIB_VERSION, \
			  (size_t) sizeof(struct jpeg_decompress_struct))

#ifdef DYNAMIC_LIBJPEG
static boolean (VARGS *qjpeg_resync_to_restart) JPP((j_decompress_ptr cinfo, int desired))									JSTATIC(jpeg_resync_to_restart);
static boolean (VARGS *qjpeg_finish_decompress) JPP((j_decompress_ptr cinfo))												JSTATIC(jpeg_finish_decompress);
static JDIMENSION (VARGS *qjpeg_read_scanlines) JPP((j_decompress_ptr cinfo, JSAMPARRAY scanlines, JDIMENSION max_lines))	JSTATIC(jpeg_read_scanlines);
static boolean (VARGS *qjpeg_start_decompress) JPP((j_decompress_ptr cinfo))												JSTATIC(jpeg_start_decompress);
static int (VARGS *qjpeg_read_header) JPP((j_decompress_ptr cinfo, boolean require_image))									JSTATIC(jpeg_read_header);
static void (VARGS *qjpeg_CreateDecompress) JPP((j_decompress_ptr cinfo, int version, size_t structsize))					JSTATIC(jpeg_CreateDecompress);
static void (VARGS *qjpeg_destroy_decompress) JPP((j_decompress_ptr cinfo))													JSTATIC(jpeg_destroy_decompress);

static struct jpeg_error_mgr * (VARGS *qjpeg_std_error) JPP((struct jpeg_error_mgr * err))									JSTATIC(jpeg_std_error);

static void (VARGS *qjpeg_finish_compress) JPP((j_compress_ptr cinfo))														JSTATIC(jpeg_finish_compress);
static JDIMENSION (VARGS *qjpeg_write_scanlines) JPP((j_compress_ptr cinfo, JSAMPARRAY scanlines, JDIMENSION num_lines))	JSTATIC(jpeg_write_scanlines);
static void (VARGS *qjpeg_write_marker) JPP((j_compress_ptr cinfo, int marker, const JOCTET *dataptr, unsigned int datalen))JSTATIC(jpeg_write_marker);
static void (VARGS *qjpeg_start_compress) JPP((j_compress_ptr cinfo, boolean write_all_tables))								JSTATIC(jpeg_start_compress);
static void (VARGS *qjpeg_set_quality) JPP((j_compress_ptr cinfo, int quality, boolean force_baseline))						JSTATIC(jpeg_set_quality);
static void (VARGS *qjpeg_set_defaults) JPP((j_compress_ptr cinfo))															JSTATIC(jpeg_set_defaults);
static void (VARGS *qjpeg_CreateCompress) JPP((j_compress_ptr cinfo, int version, size_t structsize))						JSTATIC(jpeg_CreateCompress);
static void (VARGS *qjpeg_destroy_compress) JPP((j_compress_ptr cinfo))														JSTATIC(jpeg_destroy_compress);
#endif

qboolean LibJPEG_Init(void)
{
	#ifdef DYNAMIC_LIBJPEG
	static dllfunction_t jpegfuncs[] =
	{
		{(void **) &qjpeg_resync_to_restart,		"jpeg_resync_to_restart"},
		{(void **) &qjpeg_finish_decompress,		"jpeg_finish_decompress"},
		{(void **) &qjpeg_read_scanlines,			"jpeg_read_scanlines"},
		{(void **) &qjpeg_start_decompress,			"jpeg_start_decompress"},
		{(void **) &qjpeg_read_header,				"jpeg_read_header"},
		{(void **) &qjpeg_CreateDecompress,			"jpeg_CreateDecompress"},
		{(void **) &qjpeg_destroy_decompress,		"jpeg_destroy_decompress"},

		{(void **) &qjpeg_std_error,				"jpeg_std_error"},

		{(void **) &qjpeg_finish_compress,			"jpeg_finish_compress"},
		{(void **) &qjpeg_write_scanlines,			"jpeg_write_scanlines"},
		{(void **) &qjpeg_write_marker,				"jpeg_write_marker"},
		{(void **) &qjpeg_start_compress,			"jpeg_start_compress"},
		{(void **) &qjpeg_set_quality,				"jpeg_set_quality"},
		{(void **) &qjpeg_set_defaults,				"jpeg_set_defaults"},
		{(void **) &qjpeg_CreateCompress,			"jpeg_CreateCompress"},
		{(void **) &qjpeg_destroy_compress,			"jpeg_destroy_compress"},

		{NULL, NULL}
	};

	if (!LIBJPEG_LOADED())
		libjpeg_handle = Sys_LoadLibrary("libjpeg", jpegfuncs);
#ifndef _WIN32
	if (!LIBJPEG_LOADED())
		libjpeg_handle = Sys_LoadLibrary("libjpeg"ARCH_DL_POSTFIX".8", jpegfuncs);
	if (!LIBJPEG_LOADED())
		libjpeg_handle = Sys_LoadLibrary("libjpeg"ARCH_DL_POSTFIX".62", jpegfuncs);
#endif
	#endif

	if (!LIBJPEG_LOADED())
		Con_Printf("Unable to init libjpeg\n");
	return LIBJPEG_LOADED();
}

/*begin jpeg read*/

struct my_error_mgr {
  struct jpeg_error_mgr pub;	/* "public" fields */

  jmp_buf setjmp_buffer;	/* for return to caller */
};

typedef struct my_error_mgr * my_error_ptr;

/*
 * Here's the routine that will replace the standard error_exit method:
 */

METHODDEF(void)
my_error_exit (j_common_ptr cinfo)
{
  /* cinfo->err really points to a my_error_mgr struct, so coerce pointer */
  my_error_ptr myerr = (my_error_ptr) cinfo->err;

  /* Always display the message. */
  /* We could postpone this until after returning, if we chose. */
  (*cinfo->err->output_message) (cinfo);

  /* Return control to the setjmp point */
  longjmp(myerr->setjmp_buffer, 1);
}


/*
 * Sample routine for JPEG decompression.  We assume that the source file name
 * is passed in.  We want to return 1 on success, 0 on error.
 */




/* Expanded data source object for stdio input */

typedef struct {
  struct jpeg_source_mgr pub;	/* public fields */

  qbyte * infile;		/* source stream */
  int currentpos;
  int maxlen;
  JOCTET * buffer;		/* start of buffer */
  JPEG_boolean start_of_file;	/* have we gotten any data yet? */
} my_source_mgr;

typedef my_source_mgr * my_src_ptr;

#define INPUT_BUF_SIZE  4096	/* choose an efficiently fread'able size */


METHODDEF(void)
init_source (j_decompress_ptr cinfo)
{
  my_src_ptr src = (my_src_ptr) cinfo->src;

  src->start_of_file = true;
}

METHODDEF(JPEG_boolean)
fill_input_buffer (j_decompress_ptr cinfo)
{
	my_source_mgr *src = (my_source_mgr*) cinfo->src;
	size_t nbytes;

	nbytes = src->maxlen - src->currentpos;
	if (nbytes > INPUT_BUF_SIZE)
		nbytes = INPUT_BUF_SIZE;
	memcpy(src->buffer, &src->infile[src->currentpos], nbytes);
	src->currentpos+=nbytes;

	if (nbytes <= 0)
	{
		if (src->start_of_file)	/* Treat empty input file as fatal error */
			ERREXIT(cinfo, JERR_INPUT_EMPTY);
		WARNMS(cinfo, JWRN_JPEG_EOF);
		/* Insert a fake EOI marker */
		src->buffer[0] = (JOCTET) 0xFF;
		src->buffer[1] = (JOCTET) JPEG_EOI;
		nbytes = 2;
	}

	src->pub.next_input_byte = src->buffer;
	src->pub.bytes_in_buffer = nbytes;
	src->start_of_file = false;

	return true;
}


METHODDEF(void)
skip_input_data (j_decompress_ptr cinfo, long num_bytes)
{
  my_source_mgr *src = (my_source_mgr*) cinfo->src;

  if (num_bytes > 0) {
    while (num_bytes > (long) src->pub.bytes_in_buffer) {
      num_bytes -= (long) src->pub.bytes_in_buffer;
      (void) fill_input_buffer(cinfo);
    }
    src->pub.next_input_byte += (size_t) num_bytes;
    src->pub.bytes_in_buffer -= (size_t) num_bytes;
  }
}



METHODDEF(void)
term_source (j_decompress_ptr cinfo)
{
}


#undef GLOBAL
#define GLOBAL(x) x

GLOBAL(void)
ftejpeg_mem_src (j_decompress_ptr cinfo, qbyte * infile, int maxlen)
{
  my_source_mgr *src;

  if (cinfo->src == NULL) {	/* first time for this JPEG object? */
    cinfo->src = (struct jpeg_source_mgr *)
      (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_PERMANENT,
				  sizeof(my_source_mgr));
    src = (my_source_mgr*) cinfo->src;
    src->buffer = (JOCTET *)
      (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_PERMANENT,
				  INPUT_BUF_SIZE * sizeof(JOCTET));
  }

  src = (my_source_mgr*) cinfo->src;
  src->pub.init_source = init_source;
  src->pub.fill_input_buffer = fill_input_buffer;
  src->pub.skip_input_data = skip_input_data;
  #ifdef DYNAMIC_LIBJPEG
  	src->pub.resync_to_restart = qjpeg_resync_to_restart; /* use default method */
  #else
  	src->pub.resync_to_restart = jpeg_resync_to_restart; /* use default method */
  #endif
  src->pub.term_source = term_source;
  src->infile = infile;
  src->pub.bytes_in_buffer = 0; /* forces fill_input_buffer on first read */
  src->pub.next_input_byte = NULL; /* until buffer loaded */

  src->currentpos = 0;
  src->maxlen = maxlen;
}

qbyte *ReadJPEGFile(qbyte *infile, int length, int *width, int *height)
{
	qbyte *mem=NULL, *in, *out;
	int i;

	/* This struct contains the JPEG decompression parameters and pointers to
	* working space (which is allocated as needed by the JPEG library).
	*/
	struct jpeg_decompress_struct cinfo;
	/* We use our private extension JPEG error handler.
	* Note that this struct must live as long as the main JPEG parameter
	* struct, to avoid dangling-pointer problems.
	*/
	struct my_error_mgr jerr;
	/* More stuff */
	JSAMPARRAY buffer;		/* Output row buffer */
	int size_stride;		/* physical row width in output buffer */

	memset(&cinfo, 0, sizeof(cinfo));

	if (!LIBJPEG_LOADED())
	{
		Con_DPrintf("libjpeg not available.\n");
		return NULL;
	}

	/* Step 1: allocate and initialize JPEG decompression object */

	/* We set up the normal JPEG error routines, then override error_exit. */
	#ifdef DYNAMIC_LIBJPEG
		cinfo.err = qjpeg_std_error(&jerr.pub);
	#else
		cinfo.err = jpeg_std_error(&jerr.pub);
	#endif
	jerr.pub.error_exit = my_error_exit;
	/* Establish the setjmp return context for my_error_exit to use. */
	if (setjmp(jerr.setjmp_buffer))
	{
		// If we get here, the JPEG code has signaled an error.
		Con_DPrintf("libjpeg failed to decode a file.\n");
badjpeg:
		#ifdef DYNAMIC_LIBJPEG
			qjpeg_destroy_decompress(&cinfo);
		#else
			jpeg_destroy_decompress(&cinfo);
		#endif

		if (mem)
			BZ_Free(mem);
		return NULL;
	}
	#ifdef DYNAMIC_LIBJPEG
		qjpeg_create_decompress(&cinfo);
	#else
		jpeg_create_decompress(&cinfo);
	#endif

	ftejpeg_mem_src(&cinfo, infile, length);

	#ifdef DYNAMIC_LIBJPEG
		(void) qjpeg_read_header(&cinfo, true);
	#else
		(void) jpeg_read_header(&cinfo, true);
	#endif

	#ifdef DYNAMIC_LIBJPEG
		(void) qjpeg_start_decompress(&cinfo);
	#else
		(void) jpeg_start_decompress(&cinfo);
	#endif


	if (cinfo.output_components == 0)
	{
		Con_DPrintf("No JPEG Components, not a JPEG.\n");
		goto badjpeg;
	}
	if (cinfo.output_components!=3 && cinfo.output_components != 1)
	{
		Con_DPrintf("Bad number of components in JPEG: '%d', should be '3'.\n",cinfo.output_components);
		goto badjpeg;
	}
	size_stride = cinfo.output_width * cinfo.output_components;
	/* Make a one-row-high sample array that will go away when done with image */
	buffer = (*cinfo.mem->alloc_sarray) ((j_common_ptr) &cinfo, JPOOL_IMAGE, size_stride, 1);

	out=mem=BZ_Malloc(cinfo.output_height*cinfo.output_width*4);
	memset(out, 0, cinfo.output_height*cinfo.output_width*4);

	if (cinfo.output_components == 1)
	{
		while (cinfo.output_scanline < cinfo.output_height)
		{
			#ifdef DYNAMIC_LIBJPEG
				(void) qjpeg_read_scanlines(&cinfo, buffer, 1);
			#else
				(void) jpeg_read_scanlines(&cinfo, buffer, 1);
			#endif

			in = buffer[0];
			for (i = 0; i < cinfo.output_width; i++)
			{//rgb to rgba
				*out++ = *in;
				*out++ = *in;
				*out++ = *in;
				*out++ = 255;
				in++;
			}
		}
	}
	else
	{
		while (cinfo.output_scanline < cinfo.output_height)
		{
			#ifdef DYNAMIC_LIBJPEG
				(void) qjpeg_read_scanlines(&cinfo, buffer, 1);
			#else
				(void) jpeg_read_scanlines(&cinfo, buffer, 1);
			#endif

			in = buffer[0];
			for (i = 0; i < cinfo.output_width; i++)
			{//rgb to rgba
				*out++ = *in++;
				*out++ = *in++;
				*out++ = *in++;
				*out++ = 255;
			}
		}
	}

	#ifdef DYNAMIC_LIBJPEG
		(void) qjpeg_finish_decompress(&cinfo);
	#else
		(void) jpeg_finish_decompress(&cinfo);
	#endif

	#ifdef DYNAMIC_LIBJPEG
		qjpeg_destroy_decompress(&cinfo);
	#else
		jpeg_destroy_decompress(&cinfo);
	#endif

	*width = cinfo.output_width;
	*height = cinfo.output_height;

	return mem;

}
/*end read*/
#ifndef NPFTE
/*begin write*/


#ifndef DYNAMIC_LIBJPEG
#define qjpeg_std_error			jpeg_std_error
#define qjpeg_destroy_compress	jpeg_destroy_compress
#define qjpeg_CreateCompress	jpeg_CreateCompress
#define qjpeg_set_defaults		jpeg_set_defaults
#define qjpeg_set_quality		jpeg_set_quality
#define qjpeg_write_marker		jpeg_write_marker
#define qjpeg_start_compress	jpeg_start_compress
#define qjpeg_write_scanlines	jpeg_write_scanlines
#define qjpeg_finish_compress	jpeg_finish_compress
#define qjpeg_destroy_compress	jpeg_destroy_compress
#endif
#define OUTPUT_BUF_SIZE 4096
typedef struct  {
	struct jpeg_error_mgr pub;

	jmp_buf setjmp_buffer;
} jpeg_error_mgr_wrapper;

typedef struct {
	struct jpeg_destination_mgr pub;

	vfsfile_t *vfs;


	JOCTET  buffer[OUTPUT_BUF_SIZE];		/* start of buffer */
} my_destination_mgr;

METHODDEF(void) init_destination (j_compress_ptr cinfo)
{
	my_destination_mgr *dest = (my_destination_mgr*) cinfo->dest;

	dest->pub.next_output_byte = dest->buffer;
	dest->pub.free_in_buffer = OUTPUT_BUF_SIZE;
}
METHODDEF(JPEG_boolean) empty_output_buffer (j_compress_ptr cinfo)
{
	my_destination_mgr *dest = (my_destination_mgr*) cinfo->dest;

	VFS_WRITE(dest->vfs, dest->buffer, OUTPUT_BUF_SIZE);
	dest->pub.next_output_byte = dest->buffer;
	dest->pub.free_in_buffer = OUTPUT_BUF_SIZE;

	return true;
}
METHODDEF(void) term_destination (j_compress_ptr cinfo)
{
	my_destination_mgr *dest = (my_destination_mgr*) cinfo->dest;

	VFS_WRITE(dest->vfs, dest->buffer, OUTPUT_BUF_SIZE - dest->pub.free_in_buffer);
	dest->pub.next_output_byte = dest->buffer;
	dest->pub.free_in_buffer = OUTPUT_BUF_SIZE;
}

static void ftejpeg_mem_dest (j_compress_ptr cinfo, vfsfile_t *vfs)
{
	my_destination_mgr *dest;

	if (cinfo->dest == NULL)
	{	/* first time for this JPEG object? */
		cinfo->dest = (struct jpeg_destination_mgr *)
						(*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_PERMANENT,
						sizeof(my_destination_mgr));
		dest = (my_destination_mgr*) cinfo->dest;
		//    dest->buffer = (JOCTET *)
		//      (*cinfo->mem->alloc_small) ((j_common_ptr) cinfo, JPOOL_PERMANENT,
		//				  OUTPUT_BUF_SIZE * sizeof(JOCTET));
	}

	dest = (my_destination_mgr*) cinfo->dest;
	dest->pub.init_destination = init_destination;
	dest->pub.empty_output_buffer = empty_output_buffer;
	dest->pub.term_destination = term_destination;
	dest->pub.free_in_buffer = 0; /* forces fill_input_buffer on first read */
	dest->pub.next_output_byte = NULL; /* until buffer loaded */
	dest->vfs = vfs;
}



METHODDEF(void) jpeg_error_exit (j_common_ptr cinfo)
{
	longjmp(((jpeg_error_mgr_wrapper *) cinfo->err)->setjmp_buffer, 1);
}
static qboolean screenshotJPEG(char *filename, enum fs_relative fsroot, int compression, qbyte *screendata, qintptr_t stride, int screenwidth, int screenheight, enum uploadfmt fmt, unsigned int writemeta)
{
	qbyte	*buffer;
	vfsfile_t	*outfile;
	jpeg_error_mgr_wrapper jerr;
	struct jpeg_compress_struct cinfo;
	JSAMPROW row_pointer[1];

	qbyte *rgbdata = NULL;

	//convert in-place if needed.
	//bgr->rgb may require copying out entirely for the first pixel to work properly.
	if (fmt == TF_BGRA32 || fmt == TF_BGRX32 || fmt == TF_BGR24)
	{	//byteswap and strip alpha
		size_t ps = (fmt == TF_BGR24)?3:4;
		qbyte *in=screendata, *out=rgbdata=Hunk_TempAlloc(screenwidth*screenheight*3);
		size_t y, x;
		for (y = 0; y < screenheight; y++)
		{
			for (x = 0; x < screenwidth; x++)
			{
				int r = in[2];
				int g = in[1];
				int b = in[0];
				out[0] = r;
				out[1] = g;
				out[2] = b;
				in+=ps;
				out+=3;
			}
			in-=screenwidth*ps;
			in+=stride;
		}
		fmt = TF_RGB24;
		stride = screenwidth*3;
		screendata = rgbdata;
	}
	else if (fmt == TF_RGBA32 || fmt == TF_RGBX32 || (fmt == TF_RGB24 && stride < 0))
	{	//strip alpha, no need to byteswap
		size_t ps = (fmt == TF_RGB24)?3:4;
		qbyte *in=screendata, *out=rgbdata=Hunk_TempAlloc(screenwidth*screenheight*3);
		size_t y, x;
		for (y = 0; y < screenheight; y++)
		{
			for (x = 0; x < screenwidth; x++)
			{
				int r = in[0];
				int g = in[1];
				int b = in[2];
				out[0] = r;
				out[1] = g;
				out[2] = b;
				in+=ps;
				out+=3;
			}
			in-=screenwidth*ps;
			in+=stride;
		}
		fmt = TF_RGB24;
		stride = screenwidth*3;
		screendata = rgbdata;
	}
	else if (fmt != TF_RGB24)
	{
		Con_Printf("screenshotJPEG: image format not supported\n");
		return false;
	}

	if (!LIBJPEG_LOADED())
		return false;

	if (!(outfile = FS_OpenVFS(filename, "wb", fsroot)))
	{
		FS_CreatePath (filename, fsroot);
		if (!(outfile = FS_OpenVFS(filename, "wb", fsroot)))
		{
			Con_Printf("Error opening %s\n", filename);
			return false;
		}
	}

	cinfo.err = qjpeg_std_error(&jerr.pub);
	jerr.pub.error_exit = jpeg_error_exit;
	if (setjmp(jerr.setjmp_buffer))
	{
		qjpeg_destroy_compress(&cinfo);
		VFS_CLOSE(outfile);
		FS_Remove(filename, FS_GAME);
		Con_Printf("Failed to create jpeg\n");
		return false;
	}
	qjpeg_create_compress(&cinfo);

	buffer = screendata;

	ftejpeg_mem_dest(&cinfo, outfile);
	cinfo.image_width = screenwidth;
	cinfo.image_height = screenheight;
	cinfo.input_components = 3;
	cinfo.in_color_space = JCS_RGB;
	qjpeg_set_defaults(&cinfo);
	qjpeg_set_quality (&cinfo, bound(0, compression, 100), true);
	qjpeg_start_compress(&cinfo, true);

	if (writemeta)
	{
		static const char header[] = "http://ns.adobe.com/xap/1.0/";
		char blob[8192];
		memcpy(blob, header, sizeof(header)); //MUST include the null terminator.
		GenerateXMPData(blob+sizeof(header), sizeof(blob)-sizeof(header), screenwidth, screenheight, writemeta);
		qjpeg_write_marker(&cinfo, JPEG_APP0+1, blob, sizeof(header)+strlen(blob+sizeof(header)));
	}

	while (cinfo.next_scanline < cinfo.image_height)
	{
		*row_pointer = &buffer[cinfo.next_scanline * stride];
		qjpeg_write_scanlines(&cinfo, row_pointer, 1);
	}
	qjpeg_finish_compress(&cinfo);
	VFS_CLOSE(outfile);
	qjpeg_destroy_compress(&cinfo);
	return true;
}
#endif
#endif

#ifdef IMAGEFMT_PCX
/*
==============
WritePCXfile
==============
*/
void WritePCXfile (const char *filename, enum fs_relative fsroot, qbyte *data, int width, int height,
	int rowbytes, qbyte *palette, qboolean upload) //data is 8bit.
{
	int		i, j, length;
	pcx_t	*pcx;
	qbyte		*pack;

	pcx = Hunk_TempAlloc (width*height*2+1000);
	if (pcx == NULL)
	{
		Con_Printf("WritePCXfile: not enough memory\n");
		return;
	}

	pcx->manufacturer = 0x0a;	// PCX id
	pcx->version = 5;			// 256 color
 	pcx->encoding = 1;		// uncompressed
	pcx->bits_per_pixel = 8;		// 256 color
	pcx->xmin = 0;
	pcx->ymin = 0;
	pcx->xmax = LittleShort((short)(width-1));
	pcx->ymax = LittleShort((short)(height-1));
	pcx->hres = LittleShort((short)width);
	pcx->vres = LittleShort((short)height);
	Q_memset (pcx->palette,0,sizeof(pcx->palette));
	pcx->color_planes = 1;		// chunky image
	pcx->bytes_per_line = LittleShort((short)width);
	pcx->palette_type = LittleShort(2);		// not a grey scale
	Q_memset (pcx->filler,0,sizeof(pcx->filler));

// pack the image
	pack = (qbyte *)(pcx+1);
	for (i=0 ; i<height ; i++)
	{
		for (j=0 ; j<width ; j++)
		{
			if ( (*data & 0xc0) != 0xc0)
				*pack++ = *data++;
			else
			{
				*pack++ = 0xc1;
				*pack++ = *data++;
			}
		}

		data += rowbytes - width;
	}

// write the palette
	*pack++ = 0x0c;	// palette ID qbyte
	for (i=0 ; i<768 ; i++)
		*pack++ = *palette++;

// write output file
	length = pack - (qbyte *)pcx;

	if (upload)
		CL_StartUpload((void *)pcx, length);
	else
		COM_WriteFile (filename, fsroot, pcx, length);
}

/*
============
LoadPCX
============
*/
qbyte *ReadPCXFile(qbyte *buf, int length, int *width, int *height)
{
	pcx_t	*pcx;
//	pcx_t pcxbuf;
	qbyte	*palette;
	qbyte	*pix;
	int		x, y;
	int		dataByte, runLength;
	int		count;
	qbyte *data;

	qbyte	*pcx_rgb;

	unsigned short xmin, ymin, swidth, sheight;

//
// parse the PCX file
//

	if (length < sizeof(*pcx))
		return NULL;

	pcx = (pcx_t *)buf;

	xmin = LittleShort(pcx->xmin);
	ymin = LittleShort(pcx->ymin);
	swidth = LittleShort(pcx->xmax)-xmin+1;
	sheight = LittleShort(pcx->ymax)-ymin+1;

	if (pcx->manufacturer != 0x0a
		|| pcx->version != 5
		|| pcx->encoding != 1
		|| pcx->bits_per_pixel != 8
		|| swidth >= 1024
		|| sheight >= 1024)
	{
		return NULL;
	}

	*width = swidth;
	*height = sheight;

#ifndef NPFTE
	if (r_dodgypcxfiles.value)
		palette = host_basepal;
	else
#endif
		palette = buf + length-768;

	data = (char *)(pcx+1);

	count = (swidth) * (sheight);
	pcx_rgb = BZ_Malloc( count * 4);

	for (y=0 ; y<sheight ; y++)
	{
		pix = pcx_rgb + 4*y*(swidth);
		for (x=0 ; x<swidth ; )
		{
			dataByte = *data;
			data++;

			if((dataByte & 0xC0) == 0xC0)
			{
				runLength = dataByte & 0x3F;
				if (x+runLength>swidth)
				{
					Con_Printf("corrupt pcx\n");
					BZ_Free(pcx_rgb);
					return NULL;
				}
				dataByte = *data;
				data++;
			}
			else
				runLength = 1;

			while(runLength-- > 0)
			{
				pix[0] = palette[dataByte*3];
				pix[1] = palette[dataByte*3+1];
				pix[2] = palette[dataByte*3+2];
				pix[3] = 255;
				if (dataByte == 255)
				{
					pix[0] = 0;	//linear filtering can mean transparent pixel colours are visible. black is a more neutral colour.
					pix[1] = 0;
					pix[2] = 0;
					pix[3] = 0;
				}
				pix += 4;
				x++;
			}
		}
	}

	return pcx_rgb;
}

qbyte *ReadPCXData(qbyte *buf, int length, int width, int height, qbyte *result)
{
	pcx_t	*pcx;
//	pcx_t pcxbuf;
//	qbyte	*palette;
	qbyte	*pix;
	int		x, y;
	int		dataByte, runLength;
//	int		count;
	qbyte *data;

	unsigned short xmin, ymin, swidth, sheight;

//
// parse the PCX file
//

	pcx = (pcx_t *)buf;

	xmin = LittleShort(pcx->xmin);
	ymin = LittleShort(pcx->ymin);
	swidth = LittleShort(pcx->xmax)-xmin+1;
	sheight = LittleShort(pcx->ymax)-ymin+1;

	if (pcx->manufacturer != 0x0a
		|| pcx->version != 5
		|| pcx->encoding != 1
		|| pcx->bits_per_pixel != 8)
	{
		return NULL;
	}

	if (width != swidth ||
		height > sheight)
	{
		Con_Printf("unsupported pcx size\n");
		return NULL;	//we can't feed the requester with enough info
	}

	data = (char *)(pcx+1);

	for (y=0 ; y<height ; y++)
	{
		pix = result + y*swidth;
		for (x=0 ; x<swidth ; )
		{
			dataByte = *data;
			data++;

			if((dataByte & 0xC0) == 0xC0)
			{
				runLength = dataByte & 0x3F;
				if (x+runLength>swidth)
				{
					Con_Printf("corrupt pcx\n");
					return NULL;
				}
				dataByte = *data;
				data++;
			}
			else
				runLength = 1;

			while(runLength-- > 0)
			{
				*pix++ = dataByte;
				x++;
			}
		}
	}

	return result;
}

qbyte *ReadPCXPalette(qbyte *buf, int len, qbyte *out)
{
	pcx_t	*pcx;


//
// parse the PCX file
//

	pcx = (pcx_t *)buf;

	if (pcx->manufacturer != 0x0a
		|| pcx->version != 5
		|| pcx->encoding != 1
		|| pcx->bits_per_pixel != 8
		|| LittleShort(pcx->xmax) >= 1024
		|| LittleShort(pcx->ymax) >= 1024)
	{
		return NULL;
	}

	memcpy(out, (qbyte *)pcx + len - 768, 768);

	return out;
}
#endif

#ifdef IMAGEFMT_BMP
typedef struct bmpheader_s
{
	unsigned int	SizeofBITMAPINFOHEADER;
	signed int		Width;
	signed int		Height;
	unsigned short	Planes;
	unsigned short	BitCount;
	unsigned int	Compression;
	unsigned int	ImageSize;
	signed int		TargetDeviceXRes;
	signed int		TargetDeviceYRes;
	unsigned int	NumofColorIndices;
	unsigned int	NumofImportantColorIndices;
} bmpheader_t;
typedef struct bmpheaderv4_s
{
	unsigned int	RedMask;
	unsigned int	GreenMask;
	unsigned int	BlueMask;
	unsigned int	AlphaMask;
	qbyte			ColourSpace[4];	//"Win " or "sRGB"
	qbyte			ColourSpaceCrap[12*3];
	unsigned int	Gamma[3];
} bmpheaderv4_t;

static qbyte *ReadRawBMPFile(qbyte *buf, int length, int *width, int *height, size_t OffsetofBMPBits)
{
	unsigned int i;
	bmpheader_t h;
	qbyte *data;

	memcpy(&h, buf, sizeof(h));	
	h.SizeofBITMAPINFOHEADER = LittleLong(h.SizeofBITMAPINFOHEADER);
	h.Width = LittleLong(h.Width);
	h.Height = LittleLong(h.Height);
	h.Planes = LittleShort(h.Planes);
	h.BitCount = LittleShort(h.BitCount);
	h.Compression = LittleLong(h.Compression);
	h.ImageSize = LittleLong(h.ImageSize);
	h.TargetDeviceXRes = LittleLong(h.TargetDeviceXRes);
	h.TargetDeviceYRes = LittleLong(h.TargetDeviceYRes);
	h.NumofColorIndices = LittleLong(h.NumofColorIndices);
	h.NumofImportantColorIndices = LittleLong(h.NumofImportantColorIndices);

	if (h.Compression)	//RLE? BITFIELDS (gah)?
		return NULL;

	if (!OffsetofBMPBits)
		h.Height /= 2;	//icons are weird.

	*width = h.Width;
	*height = h.Height;

	if (h.BitCount == 4)	//4 bit
	{
		int x, y;
		unsigned int *data32;
		unsigned int	pal[16];
		if (!h.NumofColorIndices)
			h.NumofColorIndices = (int)pow(2, h.BitCount);
		if (h.NumofColorIndices>16)
			return NULL;
		if (h.Width&1)
			return NULL;

		data = buf;
		data += sizeof(h);

		for (i = 0; i < h.NumofColorIndices; i++)
		{
			pal[i] = data[i*4+2] + (data[i*4+1]<<8) + (data[i*4+0]<<16) + (255u/*data[i*4+3]*/<<24);
		}

		if (OffsetofBMPBits)
			buf += OffsetofBMPBits;
		else
			buf = data+h.NumofColorIndices*4;
		data32 = BZ_Malloc(h.Width * h.Height*4);
		for (y = 0; y < h.Height; y++)
		{
			i = (h.Height-1-y) * (h.Width);
			for (x = 0; x < h.Width/2; x++)
			{
				data32[i++] = pal[buf[x]>>4];
				data32[i++] = pal[buf[x]&15];
			}
			buf += (h.Width+1)>>1;
		}

		if (!OffsetofBMPBits)
		{
			for (y = 0; y < h.Height; y++)
			{
				i = (h.Height-1-y) * (h.Width);
				for (x = 0; x < h.Width; x++)
				{
					if (buf[x>>3]&(1<<(7-(x&7))))
						data32[i] &= 0x00ffffff;
					i++;
				}
				buf += (h.Width+7)>>3;
			}
		}

		return (qbyte *)data32;
	}
	else if (h.BitCount == 8)	//8 bit
	{
		int x, y;
		unsigned int *data32;
		unsigned int	pal[256];
		if (!h.NumofColorIndices)
			h.NumofColorIndices = (int)pow(2, h.BitCount);
		if (h.NumofColorIndices>256)
			return NULL;

		data = buf;
		data += sizeof(h);

		for (i = 0; i < h.NumofColorIndices; i++)
		{
			pal[i] = data[i*4+2] + (data[i*4+1]<<8) + (data[i*4+0]<<16) + (255u/*data[i*4+3]*/<<24);
		}

		if (OffsetofBMPBits)
			buf += OffsetofBMPBits;
		else
			buf += h.SizeofBITMAPINFOHEADER + h.NumofColorIndices*4;
		data32 = BZ_Malloc(h.Width * h.Height*4);
		for (y = 0; y < h.Height; y++)
		{
			i = (h.Height-1-y) * (h.Width);
			for (x = 0; x < h.Width; x++)
			{
				data32[i] = pal[buf[x]];
				i++;
			}
			//BMP rows are 32-bit aligned.
			buf += (h.Width+3)&~3;
		}

		if (!OffsetofBMPBits)
		{
			for (y = 0; y < h.Height; y++)
			{
				i = (h.Height-1-y) * (h.Width);
				for (x = 0; x < h.Width; x++)
				{
					if (buf[x>>3]&(1<<(7-(x&7))))
						data32[i] &= 0x00ffffff;
					i++;
				}
				buf += (h.Width+7)>>3;
			}
		}

		return (qbyte *)data32;
	}
	else if (h.BitCount == 24)	//24 bit... no 16?
	{
		int x, y;
		if (OffsetofBMPBits)
			buf += OffsetofBMPBits;
		else
			buf += h.SizeofBITMAPINFOHEADER;
		data = BZ_Malloc(h.Width * h.Height*4);
		for (y = 0; y < h.Height; y++)
		{
			i = (h.Height-1-y) * (h.Width);
			for (x = 0; x < h.Width; x++)
			{
				data[i*4+0] = buf[x*3+2];
				data[i*4+1] = buf[x*3+1];
				data[i*4+2] = buf[x*3+0];
				data[i*4+3] = 255;
				i++;
			}
			buf += h.Width*3;
		}

		return data;
	}
	else if (h.BitCount == 32)
	{
		int x, y;
		if (OffsetofBMPBits)
			buf += OffsetofBMPBits;
		else
			buf += h.SizeofBITMAPINFOHEADER;
		data = BZ_Malloc(h.Width * h.Height*4);
		for (y = 0; y < h.Height; y++)
		{
			i = (h.Height-1-y) * (h.Width);
			for (x = 0; x < h.Width; x++)
			{
				data[i*4+0] = buf[x*4+2];
				data[i*4+1] = buf[x*4+1];
				data[i*4+2] = buf[x*4+0];
				data[i*4+3] = buf[x*4+3];
				i++;
			}
			buf += h.Width*4;
		}

		return data;
	}
	else
		return NULL;

	return NULL;
}

static qbyte *ReadBMPFile(qbyte *buf, int length, int *width, int *height)
{
	unsigned short Type				= buf[0] | (buf[1]<<8);
	unsigned short Size				= buf[2] | (buf[3]<<8) | (buf[4]<<16) | (buf[5]<<24);
//	unsigned short Reserved1		= buf[6] | (buf[7]<<8);
//	unsigned short Reserved2		= buf[8] | (buf[9]<<8);
	unsigned short OffsetofBMPBits	= buf[10] | (buf[11]<<8) | (buf[12]<<16) | (buf[13]<<24);
	if (Type != ('B'|('M'<<8)))
		return NULL;
	if (Size > length)
		return NULL;	//it got truncated at some point
	return ReadRawBMPFile(buf + 14, length-14, width, height, OffsetofBMPBits - 14);
}

qboolean WriteBMPFile(char *filename, enum fs_relative fsroot, qbyte *in, qintptr_t instride, int width, int height, uploadfmt_t fmt)
{
	int y;
	bmpheader_t h;
	bmpheaderv4_t h4;
	qbyte *data;
	qbyte *out;
	int outstride;
	int bits = 32;
	int extraheadersize = sizeof(h4);
	size_t fsize;
	qboolean success;

	memset(&h4, 0, sizeof(h4));
	h4.ColourSpace[0] = 'W';
	h4.ColourSpace[1] = 'i';
	h4.ColourSpace[2] = 'n';
	h4.ColourSpace[3] = ' ';
	switch(fmt)
	{
	case TF_RGBA32:
		h4.RedMask		= 0x000000ff;
		h4.GreenMask	= 0x0000ff00;
		h4.BlueMask		= 0x00ff0000;
		h4.AlphaMask	= 0xff000000;
		break;
	case TF_BGRA32:
		h4.RedMask		= 0x00ff0000;
		h4.GreenMask	= 0x0000ff00;
		h4.BlueMask		= 0x000000ff;
		h4.AlphaMask	= 0xff000000;
		break;
	case TF_RGBX32:
		h4.RedMask		= 0x000000ff;
		h4.GreenMask	= 0x0000ff00;
		h4.BlueMask		= 0x00ff0000;
		h4.AlphaMask	= 0x00000000;
		break;
	case TF_BGRX32:
		h4.RedMask		= 0x00ff0000;
		h4.GreenMask	= 0x0000ff00;
		h4.BlueMask		= 0x000000ff;
		h4.AlphaMask	= 0x00000000;
		break;
	case TF_RGB24:
		h4.RedMask		= 0x000000ff;
		h4.GreenMask	= 0x0000ff00;
		h4.BlueMask		= 0x00ff0000;
		h4.AlphaMask	= 0x00000000;
		bits = 3;
		break;
	case TF_BGR24:
		h4.RedMask		= 0x00ff0000;
		h4.GreenMask	= 0x0000ff00;
		h4.BlueMask		= 0x000000ff;
		h4.AlphaMask	= 0x00000000;
		bits = 3;
		extraheadersize = 0;
		break;

	default:
		return false;
	}


	outstride = width * (bits/8);
	outstride = (outstride+3)&~3;	//bmp pads rows to a multiple of 4 bytes.

//	h.Size = 14+sizeof(h)+extraheadersize + outstride*height;
//	h.Reserved1 = 0;
//	h.Reserved2 = 0;
//	h.OffsetofBMPBits = 2+sizeof(h)+extraheadersize;	//yes, this is misaligned.
	h.SizeofBITMAPINFOHEADER = (sizeof(h)-12)+extraheadersize;
	h.Width = width;
	h.Height = height;
	h.Planes = 1;
	h.BitCount = bits;
	h.Compression = extraheadersize?3/*BI_BITFIELDS*/:0/*BI_RGB aka BGR...*/;
	h.ImageSize = outstride*height;
	h.TargetDeviceXRes = 2835;//72DPI
	h.TargetDeviceYRes = 2835;
	h.NumofColorIndices = 0;
	h.NumofImportantColorIndices = 0;

	//bmp is bottom-up so flip it now.
	in += instride*(height-1);
	instride *= -1;

	fsize = 14+sizeof(h)+extraheadersize + outstride*height;	//size
	out = data = BZ_Malloc(fsize);
	//Type
	*out++ = 'B';
	*out++ = 'M';
	//Size
	*out++ = fsize&0xff;
	*out++ = (fsize>>8)&0xff;
	*out++ = (fsize>>16)&0xff;
	*out++ = (fsize>>24)&0xff;
	//Reserved1
	y = 0;
	*out++ = y&0xff;
	*out++ = (y>>8)&0xff;
	//Reserved1
	y = 0;
	*out++ = y&0xff;
	*out++ = (y>>8)&0xff;
	//OffsetofBMPBits
	y = 2+sizeof(h)+extraheadersize;	//yes, this is misaligned.
	*out++ = y&0xff;
	*out++ = (y>>8)&0xff;
	*out++ = (y>>16)&0xff;
	*out++ = (y>>24)&0xff;
	//bmpheader
	memcpy(out, &h, sizeof(h));
	out += sizeof(h);
	//v4 header
	memcpy(out, &h4, extraheadersize);
	out += extraheadersize;

	//data
	for (y = 0; y < height; y++)
	{
		memcpy(out, in, width * (bits/8));
		memset(out+width*(bits/8), 0, outstride-width*(bits/8));
		out += outstride;
		in += instride;
	}

	success = COM_WriteFile(filename, fsroot, data, fsize);
	BZ_Free(data);

	return success;
}

static qbyte *ReadICOFile(const char *fname, qbyte *buf, int length, int *width, int *height, uploadfmt_t *fmt)
{
	qbyte *ret;
	size_t imgcount = buf[4] | (buf[5]<<8);
	struct
	{
		qbyte bWidth;
		qbyte bHeight;
		qbyte bColorCount;
		qbyte bReserved;
		unsigned short wPlanes;
		unsigned short wBitCount;
		unsigned short dwSize_low;
		unsigned short dwSize_high;
		unsigned short dwOffset_low;
		unsigned short dwOffset_high;
	} *img = (void*)(buf+6), *bestimg = NULL;
	size_t bestpixels = 0;
	size_t bestdepth = 0;

	//always favour the png first
	for (imgcount = buf[4] | (buf[5]<<8), img = (void*)(buf+6); imgcount-->0; img++)
	{
		size_t cc = img->wBitCount;
		size_t px = (img->bWidth?img->bWidth:256) * (img->bHeight?img->bHeight:256);
		if (!cc)	//if that was omitted, try and guess it based on raw image size. this is an over estimate.
			cc = 8 * (img->dwSize_low | (img->dwSize_high<<16)) / px;

		if (!bestimg || cc > bestdepth || (cc == bestdepth && px > bestpixels))
		{
			bestimg = img;
			bestdepth = cc;
			bestpixels = px;
		}
	}

	if (bestimg)
	{
		qbyte *indata = buf + (bestimg->dwOffset_low | (bestimg->dwOffset_high<<16));
		size_t insize = (bestimg->dwSize_low | (bestimg->dwSize_high<<16));
#ifdef AVAIL_PNGLIB
		if (insize > 4 && (indata[0] == 137 && indata[1] == 'P' && indata[2] == 'N' && indata[3] == 'G') && (ret = ReadPNGFile(fname, indata, insize, width, height, fmt)))
		{
			TRACE(("dbg: Read32BitImageFile: icon png\n"));
			return ret;
		}
		else
#endif
		if ((ret = ReadRawBMPFile(indata, insize, width, height, 0)))
		{
			*fmt = PTI_RGBA8;
			TRACE(("dbg: Read32BitImageFile: icon bmp\n"));
			return ret;
		}
	}

	return NULL;
}
#endif

#ifdef IMAGEFMT_PBM
static int PBM_ParseNum(qbyte **buf, qbyte *end)
{
	qbyte token[256];
	size_t l;
	while (*buf < end)
	{
		if (**buf <= ' ')
			(*buf)++;
		else if (**buf == '#')
		{
			while (*buf < end && **buf != '\n' && **buf != '\r')
				(*buf)++;
		}
		else
			break;
	}

	for (l = 0; *buf < end && **buf > ' ';)
		if (l < countof(token))
			token[l++] = *(*buf)++;
	token[l] = 0;
	return strtol(token, NULL, 0);
}
static qbyte *ReadPBMFile(qbyte *buf, size_t len, const char *fname, int *width, int *height, uploadfmt_t *format)
{	//this isn't expected to be fast.
	qbyte *end = buf+len;
	int maxval = *width = *height = 0;
	qbyte *r, *bo;
	unsigned short *so;
	size_t l, x, y;
	float m, *fo, *fi;
	int c = buf[1];
	buf+=2;
	switch(c)
	{
	case '7': //arbitrary
		//WIDTH HEIGHT DEPTH MAXVAL TUPLTYPE ENDHDR
		return NULL;
	case '6':	//raw ppm
	case '3':	//plain ppm
		*width = PBM_ParseNum(&buf, end);
		*height = PBM_ParseNum(&buf, end);
		maxval = PBM_ParseNum(&buf, end);

		if (maxval > 255)
		{
			r = BZ_Malloc(*width**height*8);
			for(y = 0; y < *height; y++)
			for(x = 0, so=(unsigned short*)r+(*height-y-1)*4**width; x < *height; x++)
			{
				*so++ = (65535u*PBM_ParseNum(&buf, end))/maxval;
				*so++ = (65535u*PBM_ParseNum(&buf, end))/maxval;
				*so++ = (65535u*PBM_ParseNum(&buf, end))/maxval;
				*so++ = 65535u;
			}
			*format = PTI_RGBA16;
		}
		else
		{
			r = BZ_Malloc(*width**height*4);
			for(y = 0; y < *height; y++)
			for(x = 0, bo=(qbyte*)r+(*height-y-1)*4**width; x < *height; x++)
			{
				*bo++ = (255u*PBM_ParseNum(&buf, end))/maxval;
				*bo++ = (255u*PBM_ParseNum(&buf, end))/maxval;
				*bo++ = (255u*PBM_ParseNum(&buf, end))/maxval;
				*bo++ = 255u;
			}
			*format = PTI_RGBA8;
		}
		return r;
	case '5':	//raw pgm
	case '2':	//plain pgm
	case '4':	//raw pbm
	case '1':	//plain pbm
		*width = PBM_ParseNum(&buf, end);
		*height = PBM_ParseNum(&buf, end);

		if (c == '4' || c == '1')
			maxval = 1;
		else
			maxval = PBM_ParseNum(&buf, end);

		l = (size_t)*width*(size_t)*height;
		if (maxval > 255)
		{
			r = BZ_Malloc(*width**height*sizeof(*so));
			for(y = 0; y < *height; y++)
			for(x = 0, so=(unsigned short*)r+(*height-y-1)**width; x < *height; x++)
				*so++ = (65535u*PBM_ParseNum(&buf, end))/maxval;
			*format = PTI_R16;
		}
		else
		{
			r = BZ_Malloc(*width**height*sizeof(*bo));
			for(y = 0; y < *height; y++)
			for(x = 0, bo=(qbyte*)r+(*height-y-1)**width; x < *height; x++)
				*bo++ = (255u*PBM_ParseNum(&buf, end))/maxval;
			*format = PTI_R8;
		}
		return r;

	case 'F':	//rgb pfm
	case 'f':	//grey pfm
		*width = PBM_ParseNum(&buf, end);
		*height = PBM_ParseNum(&buf, end);
		m = PBM_ParseNum(&buf, end);

		if (*buf == '\n')
			buf++;
		fi = (float*)buf;

		l = (size_t)*width*(size_t)*height;
		if ((qbyte*)(fi+l*((c=='F')?3:1)) != end)
			return NULL;
		r = BZ_Malloc(l*sizeof(float) * ((c=='F')?4:1));
		if (c == 'F')
		{
			if (m < 0)
			{
				r = BZ_Malloc(*width**height*4*sizeof(float));
				for(y = 0; y < *height; y++)
				for(x = 0, fo=(float*)r+(*height-y-1)*4**width; x < *height; x++)
				{
					*fo++ = LittleFloat(*fi++);
					*fo++ = LittleFloat(*fi++);
					*fo++ = LittleFloat(*fi++);
					*fo++ = 1;
				}
			}
			else
			{
				r = BZ_Malloc(*width**height*4*sizeof(float));
				for(y = 0; y < *height; y++)
				for(x = 0, fo=(float*)r+(*height-y-1)*4**width; x < *height; x++)
				{
					*fo++ = BigFloat(*fi++);
					*fo++ = BigFloat(*fi++);
					*fo++ = BigFloat(*fi++);
					*fo++ = 1;
				}
			}
			*format = PTI_RGBA32F;
		}
		else
		{
			r = BZ_Malloc(l*sizeof(float));
			if (m < 0)
			{
				r = BZ_Malloc(*width**height*sizeof(float));
				for(y = 0; y < *height; y++)
				for(x = 0, fo=(float*)r+(*height-y-1)**width; x < *height; x++)
					*fo++ = LittleFloat(*fi++);
			}
			else
			{
				r = BZ_Malloc(*width**height*sizeof(float));
				for(y = 0; y < *height; y++)
				for(x = 0, fo=(float*)r+(*height-y-1)**width; x < *height; x++)
					*fo++ = BigFloat(*fi++);
			}
			*format = PTI_R32F;
		}
		return r; //erk?
	}
	return NULL;
}
#endif
#ifdef IMAGEFMT_HDR			//baselayer only.
//Radiance files are some weird BGRE8 hdr format that somehow managed to get reasonably well supported
static void *ReadRadianceFile(qbyte *buf, size_t len, const char *fname, int *width, int *height, uploadfmt_t *format)
{	//this isn't expected to be fast.
	qbyte *end = buf+len;
	size_t l, x, y, w, h;
	float *r, *o, m;
	qbyte rgbe[4];

	char fmt[128];
	char line[256];
	w = h = 0;
	*fmt = 0;
	while (buf < end)
	{
		l = 0;
		while(buf < end && *buf != '\n')
		{
			if (*buf == '\r' && buf[1] == '\n')
				continue;
			if (l < countof(line)-1)
				line[l++] = *buf++;
		}
		line[l] = 0;
		buf++;
		if (!strncmp(line, "FORMAT=", 7))
			Q_strncpyz(fmt, line+7, sizeof(fmt));
		if (!l)
			break;
	}
	if (strncmp(buf, "-Y ", 3))
	{
		Con_Printf("%s uses unsupported orientation\n", fname);
		return NULL;
	}
	w = strtol(buf+3, (char**)&buf, 0);
	if (strncmp(buf, " +X ", 4))
	{
		Con_Printf("%s uses unsupported orientation\n", fname);
		return NULL;
	}
	h = strtol(buf+4, (char**)&buf, 0);
	if (*buf == '\r')
		buf++;
	if (*buf++ != '\n')
		return NULL;

	if (strcmp(fmt, "32-bit_rle_rgbe"))
	{
		Con_Printf("%s uses unsupported pixel format (%s)\n", fname, fmt);
		return NULL;
	}

	r = o = BZ_Malloc(sizeof(float)*4*w*h);
	if (!r)
		return NULL;
	for (y=0; y < h; y++)
	{
		for (x=0; x < w; x++)
		{
			rgbe[0] = *buf++;
			rgbe[1] = *buf++;
			rgbe[2] = *buf++;
			rgbe[3] = *buf++;
			if (rgbe[0] == 2 && rgbe[1] == 2 && rgbe[2] < 127)
			{ //new rle logic
				Con_Printf("%s uses unsupported (new) RLE compression\n", fname);
				goto fail;
			}
			if (rgbe[0] == 1 && rgbe[1] == 1 && rgbe[2] == 1)
			{ //old rle logic
				Con_Printf("%s uses unsupported (old) RLE compression\n", fname);
				goto fail;
			}

			m = ldexp(1,rgbe[3]-136);
			*o++ = m * rgbe[0];
			*o++ = m * rgbe[1];
			*o++ = m * rgbe[2];
			*o++ = 1;
		}
	}
	*width = w;
	*height = h;
	//FIXME: should probably convert to e5bgr9 or something.
	*format = PTI_RGBA32F;
	return r;
fail:
	BZ_Free(r);
	return NULL;
}
#endif

#ifdef IMAGEFMT_PSD			//baselayer only.
// https://www.adobe.com/devnet-apps/photoshop/fileformatashtml/
struct psdctx_s
{
	qbyte *buf;
	qbyte *end;
};
static qbyte PSD_Byte(struct psdctx_s *ctx)
{
	if (ctx->buf == ctx->end)
		return 0;
	return *ctx->buf++;
}
static unsigned short PSD_UShort(struct psdctx_s *ctx)
{
	return (PSD_Byte(ctx)<<8)|PSD_Byte(ctx);
}
static unsigned int PSD_UInt(struct psdctx_s *ctx)
{
	return (PSD_Byte(ctx)<<24)|(PSD_Byte(ctx)<<16)|(PSD_Byte(ctx)<<8)|PSD_Byte(ctx);
}
static void *PSD_Block(struct psdctx_s *ctx, size_t sz)
{
	void *r;
	if (ctx->buf+sz <= ctx->end)
	{
		r = ctx->buf;
		ctx->buf += sz;
	}
	else
		r = NULL;
	return r;
}
static void *ReadPSDFile(qbyte *buf, size_t len, const char *fname, int *outwidth, int *outheight, uploadfmt_t *outformat)
{
	unsigned short ver, chans, depth, clrmode, cmp;
	unsigned int width, height, clrsize, ressize, lyrsize;
	struct psdctx_s ctx;
	size_t l, c, y;
	ctx.buf = buf;
	ctx.end = buf+len;

	/*magic	=*/ PSD_UInt(&ctx);
	ver = PSD_UShort(&ctx);
	if (ver != 1)
	{
		Con_Printf("%s unsupported .psd version\n", fname);
		return NULL;
	}
	/*reserved*/PSD_Block(&ctx, 6);
	chans   = PSD_UShort(&ctx);
	width   = PSD_UInt(&ctx);
	height  = PSD_UInt(&ctx);
	depth   = PSD_UShort(&ctx);
	clrmode = PSD_UShort(&ctx);
	clrsize = PSD_UInt(&ctx);
	/*palette =*/ PSD_Block(&ctx, clrsize);
	ressize = PSD_UInt(&ctx);
	/*resdata =*/ PSD_Block(&ctx, ressize);
	lyrsize = PSD_UInt(&ctx);
	/*lyrdata =*/ PSD_Block(&ctx, lyrsize);
	cmp		= PSD_UShort(&ctx);

	if (width <= 0 || height <= 0 || width > 16384 || height > 16384)
	{
		Con_Printf("%s too large dimensions (%u * %u)\n", fname, width, height);
		return NULL;
	}
	if (chans <= 0)
	{
		Con_Printf("%s has no colour channels\n", fname);
		return NULL;
	}
	if (clrmode == 3 && (depth == 8 || depth == 16)) //RGB
		;
	else
	{
		Con_Printf("%s not 8 or 16bpp RGB .psd image\n", fname);
		return NULL;
	}

	//the data is planer
	if (cmp == 0 || (cmp == 1 && depth==8))
	{
		if (cmp)
			PSD_Block(&ctx, 2*chans*height); //2 byte run size per plane*scanline
		if (depth == 16)
		{
			unsigned short *r, *o;
			r = o = BZ_Malloc(sizeof(*o)*4*width*height);
			for(c = 0; c < 4; c++)
			{
				o = r+c;
				if (c < chans)
				{
					for(l = 0; l < width*height; l++, o+=4)
						*o = PSD_UShort(&ctx);
				}
				else
				{	//pad colour to 0, alpha 1
					for(l = 0; l < width*height; l++, o+=4)
						*o = (c==3)?0xffff:0;
				}
			}
			*outwidth = width;
			*outheight = height;
			*outformat = PTI_RGBA16;
			return r;
		}
		else if (depth == 8)
		{
			qbyte *r, *o;
			r = o = BZ_Malloc(sizeof(*o)*4*width*height);
			for(c = 0; c < 4; c++)
			{
				o = r+c;
				if (c < chans)
				{
					if (cmp == 1)
					{
						for(y = 0; y < height; y++)
						{
							for(l = 0; l < width; )
							{
								qbyte run = PSD_Byte(&ctx), val;
								if (run < 128)
								{	//copy
									run++;
									for (; l < width && run --> 0; l++, o+=4)
										*o = PSD_Byte(&ctx);
								}
								else
								{
									run = 1-(char)run;
									val = PSD_Byte(&ctx);
									for (; l < width && run --> 0; l++, o+=4)
										*o = val;
								}
							}
						}
					}
					else for(l = 0; l < width*height; l++, o+=4)
						*o = PSD_Byte(&ctx);
				}
				else
				{	//pad colour to 0, alpha 1
					for(l = 0; l < width*height; l++, o+=4)
						*o = (c==3)?0xff:0;
				}
			}
			*outwidth = width;
			*outheight = height;
			if (chans >= 4)
				*outformat = PTI_RGBA8;
			else
				*outformat = PTI_RGBX8;
			return r;
		}
	}
	else if (cmp == 1)
	{
	}
	Con_Printf("%s unsupported compression type (%u)\n", fname, cmp);
	return NULL;
}
#endif

#ifndef NPFTE


// saturate function, stolen from jitspoe
void SaturateR8G8B8(qbyte *data, int size, float sat)
{
	int i;
	float r, g, b, v;

	if (sat > 1)
	{
		for(i=0; i < size; i+=3)
		{
			r = data[i];
			g = data[i+1];
			b = data[i+2];

			v = r * NTSC_RED + g * NTSC_GREEN + b * NTSC_BLUE;
			r = v + (r - v) * sat;
			g = v + (g - v) * sat;
			b = v + (b - v) * sat;

			// bounds check
			if (r < 0)
				r = 0;
			else if (r > 255)
				r = 255;

			if (g < 0)
				g = 0;
			else if (g > 255)
				g = 255;

			if (b < 0)
				b = 0;
			else if (b > 255)
				b = 255;

			// scale down to avoid overbright lightmaps
			v = v / (r * NTSC_RED + g * NTSC_GREEN + b * NTSC_BLUE);
			if (v > NTSC_SUM)
				v = NTSC_SUM;
			else
				v *= v;

			data[i]   = r*v;
			data[i+1] = g*v;
			data[i+2] = b*v;
		}
	}
	else // avoid bounds check for desaturation
	{
		if (sat < 0)
			sat = 0;

		for(i=0; i < size; i+=3)
		{
			r = data[i];
			g = data[i+1];
			b = data[i+2];

			v = r * NTSC_RED + g * NTSC_GREEN + b * NTSC_BLUE;

			data[i]   = v + (r - v) * sat;
			data[i+1] = v + (g - v) * sat;
			data[i+2] = v + (b - v) * sat;
		}
	}
}

void BoostGamma(qbyte *rgba, int width, int height, uploadfmt_t fmt)
{
	//note: should not be used where hardware gamma is supported.
	int i;

	switch(fmt)
	{
	case PTI_L8:
		for (i=0 ; i<width*height ; i++)
			rgba[i] = gammatable[rgba[i]];
		break;
	case PTI_LLLX8:
	case PTI_LLLA8:
	case PTI_RGBA8:
	case PTI_RGBX8:
	case PTI_RGBA8_SRGB:
	case PTI_RGBX8_SRGB:
	case PTI_BGRA8:
	case PTI_BGRX8:
	case PTI_BGRA8_SRGB:
	case PTI_BGRX8_SRGB:
		for (i=0 ; i<width*height*4 ; i+=4)
		{
			rgba[i+0] = gammatable[rgba[i+0]];
			rgba[i+1] = gammatable[rgba[i+1]];
			rgba[i+2] = gammatable[rgba[i+2]];
			//and not alpha
		}
		break;
	default:
		break;
	}
}


static void Image_LoadTexture_Failed(void *ctx, void *data, size_t a, size_t b)
{
	texid_t tex = ctx;
	tex->status = TEX_FAILED;
}
static void Image_LoadTextureMips(void *ctx, void *data, size_t a, size_t b)
{
	int i;
	texid_t tex = ctx;
	struct pendingtextureinfo *mips = data;

	//setting the dimensions here can break npot textures, so lets not do that.
//	tex->width = mips->mip[0].width;
//	tex->height = mips->mip[0].height;

	//d3d9 needs to reconfigure samplers depending on whether the data is srgb or not.
	switch(mips->encoding)
	{
	case PTI_RGBA8_SRGB:
	case PTI_RGBX8_SRGB:
	case PTI_BGRA8_SRGB:
	case PTI_BGRX8_SRGB:
	case PTI_BC1_RGB_SRGB:
	case PTI_BC1_RGBA_SRGB:
	case PTI_BC2_RGBA_SRGB:
	case PTI_BC3_RGBA_SRGB:
	case PTI_BC7_RGBA_SRGB:
	case PTI_ETC2_RGB8_SRGB:
	case PTI_ETC2_RGB8A1_SRGB:
	case PTI_ETC2_RGB8A8_SRGB:
	case PTI_ASTC_4X4_SRGB:
	case PTI_ASTC_5X4_SRGB:
	case PTI_ASTC_5X5_SRGB:
	case PTI_ASTC_6X5_SRGB:
	case PTI_ASTC_6X6_SRGB:
	case PTI_ASTC_8X5_SRGB:
	case PTI_ASTC_8X6_SRGB:
	case PTI_ASTC_10X5_SRGB:
	case PTI_ASTC_10X6_SRGB:
	case PTI_ASTC_8X8_SRGB:
	case PTI_ASTC_10X8_SRGB:
	case PTI_ASTC_10X10_SRGB:
	case PTI_ASTC_12X10_SRGB:
	case PTI_ASTC_12X12_SRGB:
		tex->flags |= IF_SRGB;
		break;
	default:
		tex->flags &= ~IF_SRGB;
		break;
	}

	if (rf->IMG_LoadTextureMips(tex, mips))
	{
		tex->format = mips->encoding;
		tex->status = TEX_LOADED;
	}
	else
	{	//failure can happen because a) lost device. b) out of device memory. c) format not supported.
		//FIXME: handle oom properly.
		tex->format = TF_INVALID;
		tex->status = TEX_FAILED;
	}

	for (i = 0; i < mips->mipcount; i++)
		if (mips->mip[i].needfree)
			BZ_Free(mips->mip[i].data);
	if (mips->extrafree)
		BZ_Free(mips->extrafree);
	BZ_Free(mips);

	//ezhud breaks without this. I assume other things will too. this is why you shouldn't depend upon querying an image's size.
	if (!strncmp(tex->ident, "gfx/", 4))
	{
		size_t lumpsize;
		qbyte lumptype;
		qpic_t *pic = W_GetLumpName(tex->ident+4, &lumpsize, &lumptype);
		if (pic && lumptype == TYP_QPIC && lumpsize == 8 + pic->width*pic->height)
		{
			tex->width = pic->width;
			tex->height = pic->height;
		}
	}
	//FIXME: check loaded wad files too.
}

#ifdef IMAGEFMT_KTX
typedef struct
{
	char magic[12];
	unsigned int endianness;

	unsigned int gltype;
	unsigned int gltypesize;
	unsigned int glformat;
	unsigned int glinternalformat;

	unsigned int glbaseinternalformat;
	unsigned int pixelwidth;
	unsigned int pixelheight;
	unsigned int pixeldepth;

	unsigned int numberofarrayelements;
	unsigned int numberoffaces;
	unsigned int numberofmipmaplevels;
	unsigned int bytesofkeyvaluedata;
} ktxheader_t;
qboolean Image_WriteKTXFile(const char *filename, enum fs_relative fsroot, struct pendingtextureinfo *mips)
{
	vfsfile_t *file;
	ktxheader_t header = {{0xAB, 0x4B, 0x54, 0x58, 0x20, 0x31, 0x31, 0xBB, 0x0D, 0x0A, 0x1A, 0x0A}, 0x04030201,
		0/*type*/, 1/*typesize*/, 0/*format*/, 0/*internalformat*/,
		0/*base*/, mips->mip[0].width, mips->mip[0].height, 0/*depth*/,
		0/*array elements*/, (mips->type==PTI_CUBEMAP)?6:1, mips->mipcount, 0/*kvdatasize*/};
	size_t mipnum;
	if (mips->type != PTI_2D && mips->type != PTI_CUBEMAP)
		return false;
	header.numberofmipmaplevels /= header.numberoffaces;

	switch(mips->encoding)
	{
	case PTI_ETC1_RGB8:			header.glinternalformat = 0x8D64/*GL_ETC1_RGB8_OES*/; break;
	case PTI_ETC2_RGB8:			header.glinternalformat = 0x9274/*GL_COMPRESSED_RGB8_ETC2*/; break;
	case PTI_ETC2_RGB8_SRGB:	header.glinternalformat = 0x9275/*GL_COMPRESSED_SRGB8_ETC2*/; break;
	case PTI_ETC2_RGB8A1:		header.glinternalformat = 0x9276/*GL_COMPRESSED_RGB8_PUNCHTHROUGH_ALPHA1_ETC2*/; break;
	case PTI_ETC2_RGB8A1_SRGB:	header.glinternalformat = 0x9277/*GL_COMPRESSED_SRGB8_PUNCHTHROUGH_ALPHA1_ETC2*/; break;
	case PTI_ETC2_RGB8A8:		header.glinternalformat = 0x9278/*GL_COMPRESSED_RGBA8_ETC2_EAC*/; break;
	case PTI_ETC2_RGB8A8_SRGB:	header.glinternalformat = 0x9279/*GL_COMPRESSED_SRGB8_ALPHA8_ETC2_EAC*/; break;
	case PTI_EAC_R11:			header.glinternalformat = 0x9270/*GL_COMPRESSED_R11_EAC*/; break;
	case PTI_EAC_R11_SNORM:		header.glinternalformat = 0x9271/*GL_COMPRESSED_SIGNED_R11_EAC*/; break;
	case PTI_EAC_RG11:			header.glinternalformat = 0x9272/*GL_COMPRESSED_RG11_EAC*/; break;
	case PTI_EAC_RG11_SNORM:	header.glinternalformat = 0x9273/*GL_COMPRESSED_SIGNED_RG11_EAC*/; break;
	case PTI_BC1_RGB:			header.glinternalformat = 0x83F0/*GL_COMPRESSED_RGB_S3TC_DXT1_EXT*/; break;
	case PTI_BC1_RGB_SRGB:		header.glinternalformat = 0x8C4C/*GL_COMPRESSED_SRGB_S3TC_DXT1_EXT*/; break;
	case PTI_BC1_RGBA:			header.glinternalformat = 0x83F1/*GL_COMPRESSED_RGBA_S3TC_DXT1_EXT*/; break;
	case PTI_BC1_RGBA_SRGB:		header.glinternalformat = 0x8C4D/*GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT1_EXT*/; break;
	case PTI_BC2_RGBA:			header.glinternalformat = 0x83F2/*GL_COMPRESSED_RGBA_S3TC_DXT3_EXT*/; break;
	case PTI_BC2_RGBA_SRGB:		header.glinternalformat = 0x8C4E/*GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT3_EXT*/; break;
	case PTI_BC3_RGBA:			header.glinternalformat = 0x83F3/*GL_COMPRESSED_RGBA_S3TC_DXT5_EXT*/; break;
	case PTI_BC3_RGBA_SRGB:		header.glinternalformat = 0x8C4F/*GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT5_EXT*/; break;
	case PTI_BC4_R8_SNORM:		header.glinternalformat = 0x8DBC/*GL_COMPRESSED_SIGNED_RED_RGTC1*/; break;
	case PTI_BC4_R8:			header.glinternalformat = 0x8DBB/*GL_COMPRESSED_RED_RGTC1*/; break;
	case PTI_BC5_RG8_SNORM:		header.glinternalformat = 0x8DBE/*GL_COMPRESSED_SIGNED_RG_RGTC2*/; break;
	case PTI_BC5_RG8:			header.glinternalformat = 0x8DBD/*GL_COMPRESSED_RG_RGTC2*/; break;
	case PTI_BC6_RGB_UFLOAT:	header.glinternalformat = 0x8E8F/*GL_COMPRESSED_RGB_BPTC_UNSIGNED_FLOAT_ARB*/; break;
	case PTI_BC6_RGB_SFLOAT:	header.glinternalformat = 0x8E8E/*GL_COMPRESSED_RGB_BPTC_SIGNED_FLOAT_ARB*/; break;
	case PTI_BC7_RGBA:			header.glinternalformat = 0x8E8C/*GL_COMPRESSED_RGBA_BPTC_UNORM_ARB*/; break;
	case PTI_BC7_RGBA_SRGB:		header.glinternalformat = 0x8E8D/*GL_COMPRESSED_SRGB_ALPHA_BPTC_UNORM_ARB*/; break;
	case PTI_ASTC_4X4:			header.glinternalformat = 0x93B0/*GL_COMPRESSED_RGBA_ASTC_4x4_KHR*/; break;
	case PTI_ASTC_5X4:			header.glinternalformat = 0x93B1/*GL_COMPRESSED_RGBA_ASTC_5x4_KHR*/; break;
	case PTI_ASTC_5X5:			header.glinternalformat = 0x93B2/*GL_COMPRESSED_RGBA_ASTC_5x5_KHR*/; break;
	case PTI_ASTC_6X5:			header.glinternalformat = 0x93B3/*GL_COMPRESSED_RGBA_ASTC_6x5_KHR*/; break;
	case PTI_ASTC_6X6:			header.glinternalformat = 0x93B4/*GL_COMPRESSED_RGBA_ASTC_6x6_KHR*/; break;
	case PTI_ASTC_8X5:			header.glinternalformat = 0x93B5/*GL_COMPRESSED_RGBA_ASTC_8x5_KHR*/; break;
	case PTI_ASTC_8X6:			header.glinternalformat = 0x93B6/*GL_COMPRESSED_RGBA_ASTC_8x6_KHR*/; break;
	case PTI_ASTC_10X5:			header.glinternalformat = 0x93B7/*GL_COMPRESSED_RGBA_ASTC_10x5_KHR*/; break;
	case PTI_ASTC_10X6:			header.glinternalformat = 0x93B8/*GL_COMPRESSED_RGBA_ASTC_10x6_KHR*/; break;
	case PTI_ASTC_8X8:			header.glinternalformat = 0x93B9/*GL_COMPRESSED_RGBA_ASTC_8x8_KHR*/; break;
	case PTI_ASTC_10X8:			header.glinternalformat = 0x93BA/*GL_COMPRESSED_RGBA_ASTC_10x8_KHR*/; break;
	case PTI_ASTC_10X10:		header.glinternalformat = 0x93BB/*GL_COMPRESSED_RGBA_ASTC_10x10_KHR*/; break;
	case PTI_ASTC_12X10:		header.glinternalformat = 0x93BC/*GL_COMPRESSED_RGBA_ASTC_12x10_KHR*/; break;
	case PTI_ASTC_12X12:		header.glinternalformat = 0x93BD/*GL_COMPRESSED_RGBA_ASTC_12x12_KHR*/; break;
	case PTI_ASTC_4X4_SRGB:		header.glinternalformat = 0x93D0/*GL_COMPRESSED_SRGB8_ALPHA8_ASTC_4x4_KHR*/; break;
	case PTI_ASTC_5X4_SRGB:		header.glinternalformat = 0x93D1/*GL_COMPRESSED_SRGB8_ALPHA8_ASTC_5x4_KHR*/; break;
	case PTI_ASTC_5X5_SRGB:		header.glinternalformat = 0x93D2/*GL_COMPRESSED_SRGB8_ALPHA8_ASTC_5x5_KHR*/; break;
	case PTI_ASTC_6X5_SRGB:		header.glinternalformat = 0x93D3/*GL_COMPRESSED_SRGB8_ALPHA8_ASTC_6x5_KHR*/; break;
	case PTI_ASTC_6X6_SRGB:		header.glinternalformat = 0x93D4/*GL_COMPRESSED_SRGB8_ALPHA8_ASTC_6x6_KHR*/; break;
	case PTI_ASTC_8X5_SRGB:		header.glinternalformat = 0x93D5/*GL_COMPRESSED_SRGB8_ALPHA8_ASTC_8x5_KHR*/; break;
	case PTI_ASTC_8X6_SRGB:		header.glinternalformat = 0x93D6/*GL_COMPRESSED_SRGB8_ALPHA8_ASTC_8x6_KHR*/; break;
	case PTI_ASTC_10X5_SRGB:	header.glinternalformat = 0x93D7/*GL_COMPRESSED_SRGB8_ALPHA8_ASTC_10x5_KHR*/; break;
	case PTI_ASTC_10X6_SRGB:	header.glinternalformat = 0x93D8/*GL_COMPRESSED_SRGB8_ALPHA8_ASTC_10x6_KHR*/; break;
	case PTI_ASTC_8X8_SRGB:		header.glinternalformat = 0x93D9/*GL_COMPRESSED_SRGB8_ALPHA8_ASTC_8x8_KHR*/; break;
	case PTI_ASTC_10X8_SRGB:	header.glinternalformat = 0x93DA/*GL_COMPRESSED_SRGB8_ALPHA8_ASTC_10x8_KHR*/; break;
	case PTI_ASTC_10X10_SRGB:	header.glinternalformat = 0x93DB/*GL_COMPRESSED_SRGB8_ALPHA8_ASTC_10x10_KHR*/; break;
	case PTI_ASTC_12X10_SRGB:	header.glinternalformat = 0x93DC/*GL_COMPRESSED_SRGB8_ALPHA8_ASTC_12x10_KHR*/; break;
	case PTI_ASTC_12X12_SRGB:	header.glinternalformat = 0x93DD/*GL_COMPRESSED_SRGB8_ALPHA8_ASTC_12x12_KHR*/; break;

	case PTI_BGRA8:				header.glinternalformat = 0x8058/*GL_RGBA8*/;				header.glbaseinternalformat = 0x1908/*GL_RGBA*/;			header.glformat = 0x80E1/*GL_BGRA*/;			header.gltype = 0x1401/*GL_UNSIGNED_BYTE*/;					header.gltypesize = 1; break;
	case PTI_RGBA8:				header.glinternalformat = 0x8058/*GL_RGBA8*/;				header.glbaseinternalformat = 0x1908/*GL_RGBA*/;			header.glformat = 0x1908/*GL_RGBA*/;			header.gltype = 0x1401/*GL_UNSIGNED_BYTE*/;					header.gltypesize = 1; break;
	case PTI_BGRA8_SRGB:		header.glinternalformat = 0x8C43/*GL_SRGB8_ALPHA8*/;		header.glbaseinternalformat = 0x1908/*GL_RGBA*/;			header.glformat = 0x80E1/*GL_BGRA*/;			header.gltype = 0x1401/*GL_UNSIGNED_BYTE*/;					header.gltypesize = 1; break;
	case PTI_RGBA8_SRGB:		header.glinternalformat = 0x8C43/*GL_SRGB8_ALPHA8*/;		header.glbaseinternalformat = 0x1908/*GL_RGBA*/;			header.glformat = 0x1908/*GL_RGBA*/;			header.gltype = 0x1401/*GL_UNSIGNED_BYTE*/;					header.gltypesize = 1; break;
	case PTI_L8:				header.glinternalformat = 0x8040/*GL_LUMINANCE8*/;			header.glbaseinternalformat = 0x1909/*GL_LUMINANCE*/;		header.glformat = 0x1909/*GL_LUMINANCE*/;		header.gltype = 0x1401/*GL_UNSIGNED_BYTE*/;					header.gltypesize = 1; break;
	case PTI_L8A8:				header.glinternalformat = 0x8045/*GL_LUMINANCE8_ALPHA8*/;	header.glbaseinternalformat = 0x190A/*GL_LUMINANCE_ALPHA*/;	header.glformat = 0x190A/*GL_LUMINANCE_ALPHA*/;	header.gltype = 0x1401/*GL_UNSIGNED_BYTE*/;					header.gltypesize = 1; break;
	case PTI_L8_SRGB:			header.glinternalformat = 0x8C47/*GL_SLUMINANCE8*/;			header.glbaseinternalformat = 0x1909/*GL_LUMINANCE*/;		header.glformat = 0x1909/*GL_LUMINANCE*/;		header.gltype = 0x1401/*GL_UNSIGNED_BYTE*/;					header.gltypesize = 1; break;
	case PTI_L8A8_SRGB:			header.glinternalformat = 0x8C45/*GL_SLUMINANCE8_ALPHA8*/;	header.glbaseinternalformat = 0x190A/*GL_LUMINANCE_ALPHA*/;	header.glformat = 0x190A/*GL_LUMINANCE_ALPHA*/;	header.gltype = 0x1401/*GL_UNSIGNED_BYTE*/;					header.gltypesize = 1; break;
	case PTI_RGB8:				header.glinternalformat = 0x8051/*GL_RGB8*/;				header.glbaseinternalformat = 0x1907/*GL_RGB*/;				header.glformat = 0x1907/*GL_RGB*/;				header.gltype = 0x1401/*GL_UNSIGNED_BYTE*/;					header.gltypesize = 1; break;
	case PTI_BGR8:				header.glinternalformat = 0x8051/*GL_RGB8*/;				header.glbaseinternalformat = 0x1907/*GL_RGB*/;				header.glformat = 0x80E0/*GL_BGR*/;				header.gltype = 0x1401/*GL_UNSIGNED_BYTE*/;					header.gltypesize = 1; break;
	case PTI_R16:				header.glinternalformat = 0x822A/*GL_R16*/;					header.glbaseinternalformat = 0x1903/*GL_RED*/;				header.glformat = 0x1903/*GL_RED*/;				header.gltype = 0x1403/*GL_UNSIGNED_SHORT*/;				header.gltypesize = 2; break;
	case PTI_RGBA16:			header.glinternalformat = 0x805B/*GL_RGBA16*/;				header.glbaseinternalformat = 0x1903/*GL_RED*/;				header.glformat = 0x1903/*GL_RED*/;				header.gltype = 0x1403/*GL_UNSIGNED_SHORT*/;				header.gltypesize = 2; break;
	case PTI_R16F:				header.glinternalformat = 0x822D/*GL_R16F*/;				header.glbaseinternalformat = 0x1903/*GL_RED*/;				header.glformat = 0x1903/*GL_RED*/;				header.gltype = 0x140B/*GL_HALF_FLOAT*/;					header.gltypesize = 2; break;
	case PTI_R32F:				header.glinternalformat = 0x822E/*GL_R32F*/;				header.glbaseinternalformat = 0x1903/*GL_RED*/;				header.glformat = 0x1903/*GL_RED*/;				header.gltype = 0x1406/*GL_FLOAT*/;							header.gltypesize = 4; break;
	case PTI_RGBA16F:			header.glinternalformat = 0x881A/*GL_RGBA16F*/;				header.glbaseinternalformat = 0x1908/*GL_RGBA*/;			header.glformat = 0x1908/*GL_RGBA*/;			header.gltype = 0x140B/*GL_HALF_FLOAT*/;					header.gltypesize = 2; break;
	case PTI_RGBA32F:			header.glinternalformat = 0x8814/*GL_RGBA32F*/;				header.glbaseinternalformat = 0x1908/*GL_RGBA*/;			header.glformat = 0x1908/*GL_RGBA*/;			header.gltype = 0x1406/*GL_FLOAT*/;							header.gltypesize = 4; break;
	case PTI_A2BGR10:			header.glinternalformat = 0x8059/*GL_RGB10_A2*/;			header.glbaseinternalformat = 0x1908/*GL_RGBA*/;			header.glformat = 0x1908/*GL_RGBA*/;			header.gltype = 0x8368/*GL_UNSIGNED_INT_2_10_10_10_REV*/;	header.gltypesize = 4; break;
	case PTI_E5BGR9:			header.glinternalformat = 0x8C3D/*GL_RGB9_E5*/;				header.glbaseinternalformat = 0x8C3D/*GL_RGB9_E5*/;			header.glformat = 0x1907/*GL_RGB*/;				header.gltype = 0x8C3E/*GL_UNSIGNED_INT_5_9_9_9_REV*/;		header.gltypesize = 4; break;
	case PTI_P8:
	case PTI_R8:				header.glinternalformat = 0x8229/*GL_R8*/;					header.glbaseinternalformat = 0x1903/*GL_RED*/;				header.glformat = 0x1903/*GL_RED*/;				header.gltype = 0x1401/*GL_UNSIGNED_BYTE*/;					header.gltypesize = 1; break;
	case PTI_RG8:				header.glinternalformat = 0x822B/*GL_RG8*/;					header.glbaseinternalformat = 0x8227/*GL_RG*/;				header.glformat = 0x8227/*GL_RG*/;				header.gltype = 0x1401/*GL_UNSIGNED_BYTE*/;					header.gltypesize = 1; break;
	case PTI_R8_SNORM:			header.glinternalformat = 0x8F94/*GL_R8_SNORM*/;			header.glbaseinternalformat = 0x1903/*GL_RED*/;				header.glformat = 0x1903/*GL_RED*/;				header.gltype = 0x1400/*GL_BYTE*/;							header.gltypesize = 1; break;
	case PTI_RG8_SNORM:			header.glinternalformat = 0x8F95/*GL_RG8_SNORM*/;			header.glbaseinternalformat = 0x8227/*GL_RG*/;				header.glformat = 0x8227/*GL_RG*/;				header.gltype = 0x1400/*GL_BYTE*/;							header.gltypesize = 1; break;
	case PTI_BGRX8:				header.glinternalformat = 0x8051/*GL_RGB8*/;				header.glbaseinternalformat = 0x1907/*GL_RGB*/;				header.glformat = 0x80E1/*GL_BGRA*/;			header.gltype = 0x1401/*GL_UNSIGNED_BYTE*/;					header.gltypesize = 1; break;
	case PTI_RGBX8:				header.glinternalformat = 0x8051/*GL_RGB8*/;				header.glbaseinternalformat = 0x1907/*GL_RGB*/;				header.glformat = 0x1908/*GL_RGBA*/;			header.gltype = 0x1401/*GL_UNSIGNED_BYTE*/;					header.gltypesize = 1; break;
	case PTI_BGRX8_SRGB:		header.glinternalformat = 0x8C41/*GL_SRGB8*/;				header.glbaseinternalformat = 0x1908/*GL_RGBA*/;			header.glformat = 0x80E1/*GL_BGRA*/;			header.gltype = 0x1401/*GL_UNSIGNED_BYTE*/;					header.gltypesize = 1; break;
	case PTI_RGBX8_SRGB:		header.glinternalformat = 0x8C41/*GL_SRGB8*/;				header.glbaseinternalformat = 0x1908/*GL_RGBA*/;			header.glformat = 0x1908/*GL_RGBA*/;			header.gltype = 0x1401/*GL_UNSIGNED_BYTE*/;					header.gltypesize = 1; break;
	case PTI_RGB565:			header.glinternalformat = 0x8D62/*GL_RGB565*/;				header.glbaseinternalformat = 0x1907/*GL_RGB*/;				header.glformat = 0x1907/*GL_RGB*/;				header.gltype = 0x8363/*GL_UNSIGNED_SHORT_5_6_5*/;			header.gltypesize = 2; break;
	case PTI_RGBA4444:			header.glinternalformat = 0x8056/*GL_RGBA4*/;				header.glbaseinternalformat = 0x1908/*GL_RGBA*/;			header.glformat = 0x1908/*GL_RGBA*/;			header.gltype = 0x8033/*GL_UNSIGNED_SHORT_4_4_4_4*/;		header.gltypesize = 2; break;
	case PTI_ARGB4444:			header.glinternalformat = 0x8056/*GL_RGBA4*/;				header.glbaseinternalformat = 0x1908/*GL_RGBA*/;			header.glformat = 0x80E1/*GL_BGRA*/;			header.gltype = 0x8365/*GL_UNSIGNED_SHORT_4_4_4_4_REV*/;	header.gltypesize = 2; break;
	case PTI_RGBA5551:			header.glinternalformat = 0x8057/*GL_RGB5_A1*/;				header.glbaseinternalformat = 0x1908/*GL_RGBA*/;			header.glformat = 0x1908/*GL_RGBA*/;			header.gltype = 0x8034/*GL_UNSIGNED_SHORT_5_5_5_1*/;		header.gltypesize = 2; break;
	case PTI_ARGB1555:			header.glinternalformat = 0x8057/*GL_RGB5_A1*/;				header.glbaseinternalformat = 0x1908/*GL_RGBA*/;			header.glformat = 0x80E1/*GL_BGRA*/;			header.gltype = 0x8366/*GL_UNSIGNED_SHORT_1_5_5_5_REV*/;	header.gltypesize = 2; break;
	case PTI_DEPTH16:			header.glinternalformat = 0x81A5/*GL_DEPTH_COMPONENT16*/;	header.glbaseinternalformat = 0x1902/*GL_DEPTH_COMPONENT*/;	header.glformat = 0x1902/*GL_DEPTH_COMPONENT*/;	header.gltype = 0x1403/*GL_UNSIGNED_SHORT*/;				header.gltypesize = 2; break;
	case PTI_DEPTH24:			header.glinternalformat = 0x81A6/*GL_DEPTH_COMPONENT24*/;	header.glbaseinternalformat = 0x1902/*GL_DEPTH_COMPONENT*/;	header.glformat = 0x1902/*GL_DEPTH_COMPONENT*/;	header.gltype = 0x1405/*GL_UNSIGNED_INT*/;					header.gltypesize = 4; break;
	case PTI_DEPTH32:			header.glinternalformat = 0x81A7/*GL_DEPTH_COMPONENT32*/;	header.glbaseinternalformat = 0x1902/*GL_DEPTH_COMPONENT*/;	header.glformat = 0x1902/*GL_DEPTH_COMPONENT*/;	header.gltype = 0x1406/*GL_FLOAT*/;							header.gltypesize = 4; break;
	case PTI_DEPTH24_8:			header.glinternalformat = 0x88F0/*GL_DEPTH24_STENCIL8*/;	header.glbaseinternalformat = 0x84F9/*GL_DEPTH_STENCIL*/;	header.glformat = 0x84F9/*GL_DEPTH_STENCIL*/;	header.gltype = 0x84FA/*GL_UNSIGNED_INT_24_8*/;				header.gltypesize = 4; break;

#ifdef FTE_TARGET_WEB
	case PTI_WHOLEFILE:
#endif
	case PTI_EMULATED:
	case PTI_MAX:
		return false;

//	default:
//		return;
	}

	if (strchr(filename, '*') || strchr(filename, ':'))
		return false;

	file = FS_OpenVFS(filename, "wb", fsroot);
	if (!file)
		return false;
	VFS_WRITE(file, &header, sizeof(header));

	for (mipnum = 0; mipnum < mips->mipcount; mipnum++)
	{
		unsigned int pad = 0;
		unsigned int sz = mips->mip[mipnum].datasize;
		if (!(mipnum % header.numberoffaces))
			VFS_WRITE(file, &sz, 4);

		VFS_WRITE(file, mips->mip[mipnum].data, sz);
		if ((sz & 3) && mips->type == PTI_CUBEMAP)
			VFS_WRITE(file, &pad, 4-(sz&3));
	}

	VFS_CLOSE(file);
	return true;
}
static struct pendingtextureinfo *Image_ReadKTXFile(unsigned int flags, const char *fname, qbyte *filedata, size_t filesize)
{
	static const char magic[12] = {0xAB, 0x4B, 0x54, 0x58, 0x20, 0x31, 0x31, 0xBB, 0x0D, 0x0A, 0x1A, 0x0A};
	ktxheader_t *header;
	int nummips;
	int mipnum;
	int face;
	int datasize;
	unsigned int w, h, d;
	struct pendingtextureinfo *mips;
	int encoding = TF_INVALID;
	qbyte *in;
	qbyte *fileend = filedata + filesize;

	unsigned int blockwidth, blockheight, blockbytes;

	if (memcmp(filedata, magic, sizeof(magic)))
		return NULL;	//not a ktx file

	header = (ktxheader_t*)filedata;
	nummips = header->numberofmipmaplevels;
	if (nummips < 1)
		nummips = 1;

	if (header->numberofarrayelements != 0)
		return NULL;	//don't support array textures
	if (header->numberoffaces == 1)
		;	//non-cubemap
	else if (header->numberoffaces == 6)
	{
		if (header->pixeldepth != 0)
			return NULL;
//		if (header->numberofmipmaplevels != 1)
//			return false;	//only allow cubemaps that have no mips
	}
	else
		return NULL;	//don't allow weird cubemaps
	if (header->pixeldepth && header->pixelwidth != header->pixeldepth && header->pixelheight != header->pixeldepth)
		return NULL;	//we only support 3d textures where width+height+depth are the same. too lazy to change it now.

	/*FIXME: validate format+type for non-compressed formats*/
	switch(header->glinternalformat)
	{
	case 0x8D64/*GL_ETC1_RGB8_OES*/:							encoding = PTI_ETC1_RGB8;			break;
	case 0x9270/*GL_COMPRESSED_R11_EAC*/:						encoding = PTI_EAC_R11;				break;
	case 0x9271/*GL_COMPRESSED_SIGNED_R11_EAC*/:				encoding = PTI_EAC_R11_SNORM;		break;
	case 0x9272/*GL_COMPRESSED_RG11_EAC*/:						encoding = PTI_EAC_RG11;			break;
	case 0x9273/*GL_COMPRESSED_SIGNED_RG11_EAC*/:				encoding = PTI_EAC_RG11_SNORM;		break;
	case 0x9274/*GL_COMPRESSED_RGB8_ETC2*/:						encoding = PTI_ETC2_RGB8;			break;
	case 0x9275/*GL_COMPRESSED_SRGB8_ETC2*/:					encoding = PTI_ETC2_RGB8_SRGB;		break;
	case 0x9276/*GL_COMPRESSED_RGB8_PUNCHTHROUGH_ALPHA1_ETC2*/:	encoding = PTI_ETC2_RGB8A1;			break;
	case 0x9277/*GL_COMPRESSED_SRGB8_PUNCHTHROUGH_ALPHA1_ETC2*/:encoding = PTI_ETC2_RGB8A1_SRGB;	break;
	case 0x9278/*GL_COMPRESSED_RGBA8_ETC2_EAC*/:				encoding = PTI_ETC2_RGB8A8;			break;
	case 0x9279/*GL_COMPRESSED_SRGB8_ALPHA8_ETC2_EAC*/:			encoding = PTI_ETC2_RGB8A8_SRGB;	break;
	case 0x83F0/*GL_COMPRESSED_RGB_S3TC_DXT1_EXT*/:				encoding = PTI_BC1_RGB;				break;
	case 0x8C4C/*GL_COMPRESSED_SRGB_S3TC_DXT1_EXT*/:			encoding = PTI_BC1_RGB_SRGB;		break;
	case 0x83F1/*GL_COMPRESSED_RGBA_S3TC_DXT1_EXT*/:			encoding = PTI_BC1_RGBA;			break;
	case 0x8C4D/*GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT1_EXT*/:		encoding = PTI_BC1_RGBA_SRGB;		break;
	case 0x83F2/*GL_COMPRESSED_RGBA_S3TC_DXT3_EXT*/:			encoding = PTI_BC2_RGBA;			break;
	case 0x8C4E/*GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT3_EXT*/:		encoding = PTI_BC2_RGBA_SRGB;		break;
	case 0x83F3/*GL_COMPRESSED_RGBA_S3TC_DXT5_EXT*/:			encoding = PTI_BC3_RGBA;			break;
	case 0x8C4F/*GL_COMPRESSED_SRGB_ALPHA_S3TC_DXT5_EXT*/:		encoding = PTI_BC3_RGBA_SRGB;		break;
	case 0x8DBC/*GL_COMPRESSED_SIGNED_RED_RGTC1*/:				encoding = PTI_BC4_R8_SNORM;		break;
	case 0x8DBB/*GL_COMPRESSED_RED_RGTC1*/:						encoding = PTI_BC4_R8;				break;
	case 0x8DBE/*GL_COMPRESSED_SIGNED_RG_RGTC2*/:				encoding = PTI_BC5_RG8_SNORM;		break;
	case 0x8DBD/*GL_COMPRESSED_RG_RGTC2*/:						encoding = PTI_BC5_RG8;				break;
	case 0x8E8F/*GL_COMPRESSED_RGB_BPTC_UNSIGNED_FLOAT_ARB*/:	encoding = PTI_BC6_RGB_UFLOAT;		break;
	case 0x8E8E/*GL_COMPRESSED_RGB_BPTC_SIGNED_FLOAT_ARB*/:		encoding = PTI_BC6_RGB_SFLOAT;		break;
	case 0x8E8C/*GL_COMPRESSED_RGBA_BPTC_UNORM_ARB*/:			encoding = PTI_BC7_RGBA;			break;
	case 0x8E8D/*GL_COMPRESSED_SRGB_ALPHA_BPTC_UNORM_ARB*/:		encoding = PTI_BC7_RGBA_SRGB;		break;
	case 0x93B0/*GL_COMPRESSED_RGBA_ASTC_4x4_KHR*/:				encoding = PTI_ASTC_4X4;			break;
	case 0x93B1/*GL_COMPRESSED_RGBA_ASTC_5x4_KHR*/:				encoding = PTI_ASTC_5X4;			break;
	case 0x93B2/*GL_COMPRESSED_RGBA_ASTC_5x5_KHR*/:				encoding = PTI_ASTC_5X5;			break;
	case 0x93B3/*GL_COMPRESSED_RGBA_ASTC_6x5_KHR*/:				encoding = PTI_ASTC_6X5;			break;
	case 0x93B4/*GL_COMPRESSED_RGBA_ASTC_6x6_KHR*/:				encoding = PTI_ASTC_6X6;			break;
	case 0x93B5/*GL_COMPRESSED_RGBA_ASTC_8x5_KHR*/:				encoding = PTI_ASTC_8X5;			break;
	case 0x93B6/*GL_COMPRESSED_RGBA_ASTC_8x6_KHR*/:				encoding = PTI_ASTC_8X6;			break;
	case 0x93B7/*GL_COMPRESSED_RGBA_ASTC_10x5_KHR*/:			encoding = PTI_ASTC_10X5;			break;
	case 0x93B8/*GL_COMPRESSED_RGBA_ASTC_10x6_KHR*/:			encoding = PTI_ASTC_10X6;			break;
	case 0x93B9/*GL_COMPRESSED_RGBA_ASTC_8x8_KHR*/:				encoding = PTI_ASTC_8X8;			break;
	case 0x93BA/*GL_COMPRESSED_RGBA_ASTC_10x8_KHR*/:			encoding = PTI_ASTC_10X8;			break;
	case 0x93BB/*GL_COMPRESSED_RGBA_ASTC_10x10_KHR*/:			encoding = PTI_ASTC_10X10;			break;
	case 0x93BC/*GL_COMPRESSED_RGBA_ASTC_12x10_KHR*/:			encoding = PTI_ASTC_12X10;			break;
	case 0x93BD/*GL_COMPRESSED_RGBA_ASTC_12x12_KHR*/:			encoding = PTI_ASTC_12X12;			break;
	case 0x93D0/*GL_COMPRESSED_SRGB8_ALPHA8_ASTC_4x4_KHR*/:		encoding = PTI_ASTC_4X4_SRGB;		break;
	case 0x93D1/*GL_COMPRESSED_SRGB8_ALPHA8_ASTC_5x4_KHR*/:		encoding = PTI_ASTC_5X4_SRGB;		break;
	case 0x93D2/*GL_COMPRESSED_SRGB8_ALPHA8_ASTC_5x5_KHR*/:		encoding = PTI_ASTC_5X5_SRGB;		break;
	case 0x93D3/*GL_COMPRESSED_SRGB8_ALPHA8_ASTC_6x5_KHR*/:		encoding = PTI_ASTC_6X5_SRGB;		break;
	case 0x93D4/*GL_COMPRESSED_SRGB8_ALPHA8_ASTC_6x6_KHR*/:		encoding = PTI_ASTC_6X6_SRGB;		break;
	case 0x93D5/*GL_COMPRESSED_SRGB8_ALPHA8_ASTC_8x5_KHR*/:		encoding = PTI_ASTC_8X5_SRGB;		break;
	case 0x93D6/*GL_COMPRESSED_SRGB8_ALPHA8_ASTC_8x6_KHR*/:		encoding = PTI_ASTC_8X6_SRGB;		break;
	case 0x93D7/*GL_COMPRESSED_SRGB8_ALPHA8_ASTC_10x5_KHR*/:	encoding = PTI_ASTC_10X5_SRGB;		break;
	case 0x93D8/*GL_COMPRESSED_SRGB8_ALPHA8_ASTC_10x6_KHR*/:	encoding = PTI_ASTC_10X6_SRGB;		break;
	case 0x93D9/*GL_COMPRESSED_SRGB8_ALPHA8_ASTC_8x8_KHR*/:		encoding = PTI_ASTC_8X8_SRGB;		break;
	case 0x93DA/*GL_COMPRESSED_SRGB8_ALPHA8_ASTC_10x8_KHR*/:	encoding = PTI_ASTC_10X8_SRGB;		break;
	case 0x93DB/*GL_COMPRESSED_SRGB8_ALPHA8_ASTC_10x10_KHR*/:	encoding = PTI_ASTC_10X10_SRGB;		break;
	case 0x93DC/*GL_COMPRESSED_SRGB8_ALPHA8_ASTC_12x10_KHR*/:	encoding = PTI_ASTC_12X10_SRGB;		break;
	case 0x93DD/*GL_COMPRESSED_SRGB8_ALPHA8_ASTC_12x12_KHR*/:	encoding = PTI_ASTC_12X12_SRGB;		break;
	case 0x80E1/*GL_BGRA_EXT*/:									encoding = PTI_BGRA8;				break;	//not even an internal format
	case 0x1908/*GL_RGBA*/:
	case 0x8058/*GL_RGBA8*/:									encoding = (header->glformat==0x80E1/*GL_BGRA*/)?PTI_BGRA8:PTI_RGBA8;			break;	//unsized types shouldn't really be here
	case 0x805B/*GL_RGBA16*/:									encoding = PTI_RGBA16;				break;
	case 0x8C43/*GL_SRGB8_ALPHA8*/:								encoding = (header->glformat==0x80E1/*GL_BGRA*/)?PTI_BGRA8_SRGB:PTI_RGBA8_SRGB;	break;
	case 0x8040/*GL_LUMINANCE8*/:								encoding = PTI_L8;					break;
	case 0x8045/*GL_LUMINANCE8_ALPHA8*/:						encoding = PTI_L8A8;				break;
	case 0x881A/*GL_RGBA16F_ARB*/:								encoding = PTI_RGBA16F;				break;
	case 0x8814/*GL_RGBA32F_ARB*/:								encoding = PTI_RGBA32F;				break;
	case 0x8059/*GL_RGB10_A2*/:									encoding = PTI_A2BGR10;				break;
	case 0x8229/*GL_R8*/:										encoding = PTI_R8;					break;
	case 0x822A/*GL_R16*/:										encoding = PTI_R16;					break;
	case 0x822B/*GL_RG8*/:										encoding = PTI_RG8;					break;
	case 0x8F94/*GL_R8_SNORM*/:									encoding = PTI_R8_SNORM;			break;
	case 0x8F95/*GL_RG8_SNORM*/:								encoding = PTI_RG8_SNORM;			break;
	case 0x81A5/*GL_DEPTH_COMPONENT16*/:						encoding = PTI_DEPTH16;				break;
	case 0x81A6/*GL_DEPTH_COMPONENT24*/:						encoding = PTI_DEPTH24;				break;
	case 0x81A7/*GL_DEPTH_COMPONENT32*/:						encoding = PTI_DEPTH32;				break;
	case 0x88F0/*GL_DEPTH24_STENCIL8*/:							encoding = PTI_DEPTH24_8;			break;
	case 0x822D/*GL_R16F*/:										encoding = PTI_R16F;				break;
	case 0x822E/*GL_R32F*/:										encoding = PTI_R32F;				break;

	case 0x8C40/*GL_SRGB*/:
	case 0x8C41/*GL_SRGB8*/:
		if (header->glformat==0x80E1/*GL_BGRA*/)
			encoding = PTI_BGRX8_SRGB;
		else if (header->glformat==0x1908/*GL_RGBA*/)
			encoding = PTI_RGBX8_SRGB;
		break;

	case 0x1907/*GL_RGB*/:	//invalid sized format. treat as GL_RGB8, and do weird checks.
	case 0x8051/*GL_RGB8*/:	//other sized RGB formats are treated based upon the data format rather than the sized format, in case they were meant to be converted by the driver...
	case 0x8C3D/*GL_RGB9_E5*/:
	case 0x8D62/*GL_RGB565*/:
		if (header->glformat == 0x80E0/*GL_BGR*/)
			encoding = PTI_BGR8;
		else if (header->glformat == 0x80E1/*GL_BGRA*/)
			encoding = PTI_BGRX8;
		else if (header->glformat == 0x1907/*GL_RGB*/)
		{
			if (header->gltype == 0x8C3E/*GL_UNSIGNED_INT_5_9_9_9_REV*/)
				encoding = PTI_E5BGR9;
			else if (header->gltype == 0x8363/*GL_UNSIGNED_SHORT_5_6_5*/)
				encoding = PTI_RGB565;
			else
				encoding = PTI_RGB8;
		}
		else if (header->glformat == 0x1908/*GL_RGBA*/)
			encoding = PTI_RGBX8;
		else
			encoding = PTI_RGB8;
		break;
	case 0x8056/*GL_RGBA4*/:
	case 0x8057/*GL_RGB5_A1*/:
		if (header->glformat == 0x1908/*GL_RGBA*/ && header->gltype == 0x8034/*GL_UNSIGNED_SHORT_5_5_5_1*/)
			encoding = PTI_RGBA5551;
		else if (header->glformat == 0x80E1/*GL_BGRA*/ && header->gltype == 0x8366/*GL_UNSIGNED_SHORT_1_5_5_5_REV*/)
			encoding = PTI_ARGB1555;
		else if (header->glformat == 0x1908/*GL_RGBA*/ && header->gltype == 0x8033/*GL_UNSIGNED_SHORT_4_4_4_4*/)
			encoding = PTI_RGBA4444;
		else if (header->glformat == 0x80E1/*GL_BGRA*/ && header->gltype == 0x8365/*GL_UNSIGNED_SHORT_4_4_4_4_REV*/)
			encoding = PTI_ARGB4444;
		break;
	}
	if (encoding == TF_INVALID)
	{
		Con_Printf("Unsupported ktx internalformat %x in %s\n", header->glinternalformat, fname);
		return NULL;
	}

//	if (!sh_config.texfmt[encoding])
//	{
//		Con_Printf("KTX %s: encoding %x not supported on this system\n", fname, header->glinternalformat);
//		return false;
//	}

	mips = Z_Malloc(sizeof(*mips));
	mips->mipcount = 0;
	if (header->pixeldepth)
		mips->type = PTI_3D;
	else if (header->numberoffaces==6)
	{
		if (header->numberofarrayelements)
		{
			header->pixeldepth = header->numberofarrayelements*6;
			mips->type = PTI_CUBEMAP_ARRAY;
		}
		else
			mips->type = PTI_CUBEMAP;
	}
	else
	{
		if (header->numberofarrayelements)
		{
			header->pixeldepth = header->numberofarrayelements;
			mips->type = PTI_2D_ARRAY;
		}
		else
		{
			header->pixeldepth = 1;
			mips->type = PTI_2D;
		}
	}
	mips->extrafree = filedata;
	mips->encoding = encoding;

	filedata += sizeof(*header);			//skip the header...
	filedata += header->bytesofkeyvaluedata;	//skip the keyvalue stuff

	if (nummips * header->numberoffaces > countof(mips->mip))
		nummips = countof(mips->mip) / header->numberoffaces;

	Image_BlockSizeForEncoding(encoding, &blockbytes, &blockwidth, &blockheight);

	w = header->pixelwidth;
	h = header->pixelheight;
	d = header->pixeldepth;

	//fixme: if (w+blockwidth-1)/blockwidth)*blockbytes MUST be a multiple of 4.
	//we need to de-pad it otherwise.

	for (mipnum = 0; mipnum < nummips; mipnum++)
	{
		datasize = *(int*)filedata;
		filedata += 4;

		if (datasize != blockbytes * ((w+blockwidth-1)/blockwidth) * ((h+blockheight-1)/blockheight))
		{
			Con_Printf("%s: mip %i does not match expected size\n", fname, mipnum);
			break;
		}

		if (filedata + datasize*header->numberoffaces > fileend)
		{
			Con_Printf("%s: truncation at mip %i\n", fname, mipnum);
			break;
		}

		for (face = 0; face < header->numberoffaces; face++)
		{
			if (mips->mipcount >= countof(mips->mip))
				break;
			mips->mip[mips->mipcount].data = in = filedata;
			mips->mip[mips->mipcount].datasize = datasize;
			mips->mip[mips->mipcount].width = w;
			mips->mip[mips->mipcount].height = h;
			mips->mip[mips->mipcount].depth = d;
			mips->mipcount++;

			filedata += datasize;
			if ((datasize & 3) && mips->type == PTI_CUBEMAP)
				filedata += 4-(datasize&3);
		}
		w = max(1, w>>1);
		h = max(1, h>>1);
		if (mips->type == PTI_3D)
			d = max(1, d>>1);
	}

	if (!mips->mipcount)
	{
		Z_Free(mips);
		return NULL;
	}

	return mips;
}
#endif

#ifdef IMAGEFMT_PKM
static struct pendingtextureinfo *Image_ReadPKMFile(unsigned int flags, const char *fname, qbyte *filedata, size_t filesize)
{
	struct pendingtextureinfo *mips;
	unsigned int encoding, blockbytes, blockwidth, blockheight;
	unsigned short ver, dfmt;
	unsigned short datawidth, dataheight;
	unsigned short imgwidth, imgheight;
	if (filedata[0] != 'P' || filedata[1] != 'K' || filedata[2] != 'M' || filedata[3] != ' ')
		return NULL;
	ver = (filedata[4]<<8) | filedata[5];
	dfmt = (filedata[6]<<8) | filedata[7];
	datawidth = (filedata[8]<<8) | filedata[9];
	dataheight = (filedata[10]<<8) | filedata[11];
	imgwidth = (filedata[12]<<8) | filedata[13];
	imgheight = (filedata[14]<<8) | filedata[15];
	if (((imgwidth+3)&~3)!=datawidth || ((imgheight+3)&~3)!=dataheight)
		return NULL;	//these are all 4*4 blocks.
	if (ver == ((1<<8)|0) && ver == ((2<<8)|0))
	{
		if (dfmt == 0)	//should only be in v1
			encoding = PTI_ETC1_RGB8;
		//following should only be in v2, but we don't care.
		else if (dfmt == 1)
			encoding = PTI_ETC2_RGB8;
		else if (dfmt == 2)
			return NULL;	//'old' rgba8 format that's not supported.
		else if (dfmt == 3)
			encoding = PTI_ETC2_RGB8A8;
		else if (dfmt == 4)
			encoding = PTI_ETC2_RGB8A1;
		else if (dfmt == 5)
			encoding = PTI_EAC_R11;
		else if (dfmt == 6)
			encoding = PTI_EAC_RG11;
		else if (dfmt == 7)
			encoding = PTI_EAC_R11_SNORM;
		else if (dfmt == 8)
			encoding = PTI_EAC_RG11_SNORM;
		else if (dfmt == 9)
			encoding = PTI_ETC2_RGB8_SRGB;	//srgb
		else if (dfmt == 10)
			encoding = PTI_ETC2_RGB8A8_SRGB;	//srgb
		else if (dfmt == 11)
			encoding = PTI_ETC2_RGB8A1_SRGB;	//srgb
		else
			return NULL;	//unknown/unsupported
	}
	else
		return NULL;

	Image_BlockSizeForEncoding(encoding, &blockbytes, &blockwidth, &blockheight);
	if (16+((datawidth+blockwidth-1)/blockwidth)*((dataheight+blockheight-1)/blockheight)*blockbytes != filesize)
		return NULL;	//err, not the right size!

	mips = Z_Malloc(sizeof(*mips));
	mips->mipcount = 1;	//this format doesn't support mipmaps. so there's only one level.
	mips->type = PTI_2D;
	mips->extrafree = filedata;
	mips->encoding = encoding;
	mips->mip[0].data = filedata+16;
	mips->mip[0].datasize = filesize-16;
	mips->mip[0].width = imgwidth;
	mips->mip[0].height = imgheight;
	mips->mip[0].depth = 1;
	mips->mip[0].needfree = false;
	return mips;
}
#endif

#ifdef IMAGEFMT_DDS
typedef struct {
	unsigned int dwSize;
	unsigned int dwFlags;
	unsigned int dwFourCC;

	unsigned int unk[5];
} ddspixelformat;

typedef struct {
	unsigned int dwSize;
	unsigned int dwFlags;
	unsigned int dwHeight;
	unsigned int dwWidth;
	unsigned int dwPitchOrLinearSize;
	unsigned int dwDepth;
	unsigned int dwMipMapCount;
	unsigned int dwReserved1[11];
	ddspixelformat ddpfPixelFormat;
	unsigned int ddsCaps[4];
	unsigned int dwReserved2;
} ddsheader;
typedef struct {
	unsigned int dxgiformat;
	unsigned int resourcetype; //0=unknown, 1=buffer, 2=1d, 3=2d, 4=3d
	unsigned int miscflag;	//singular... yeah. 4=cubemap.
	unsigned int arraysize;
	unsigned int miscflags2;
} dds10header_t;

static struct pendingtextureinfo *Image_ReadDDSFile(unsigned int flags, const char *fname, qbyte *filedata, size_t filesize)
{
	int nummips;
	int mipnum;
	int datasize;
//	int pad;
	unsigned int w, h;
	unsigned int blockwidth, blockheight, blockbytes;
	struct pendingtextureinfo *mips;
	int encoding;
	int layers = 1, layer;

	ddsheader fmtheader;
	dds10header_t fmt10header;

	if (*(int*)filedata != (('D'<<0)|('D'<<8)|('S'<<16)|(' '<<24)))
		return NULL;

	memcpy(&fmtheader, filedata+4, sizeof(fmtheader));
	if (fmtheader.dwSize != sizeof(fmtheader))
		return NULL;	//corrupt/different version
	memset(&fmt10header, 0, sizeof(fmt10header));
	fmt10header.arraysize = 1;

	nummips = fmtheader.dwMipMapCount;
	if (nummips < 1)
		nummips = 1;

	if (*(int*)&fmtheader.ddpfPixelFormat.dwFourCC == (('D'<<0)|('X'<<8)|('T'<<16)|('1'<<24)))
	{
		encoding = PTI_BC1_RGBA;	//alpha or not? Assume yes, and let the drivers decide.
//		pad = 8;
	}
	else if (*(int*)&fmtheader.ddpfPixelFormat.dwFourCC == (('D'<<0)|('X'<<8)|('T'<<16)|('2'<<24)))	//dx3 with premultiplied alpha
	{
//		if (!(tex->flags & IF_PREMULTIPLYALPHA))
//			return false;
		encoding = PTI_BC2_RGBA;
//		pad = 8;
	}
	else if (*(int*)&fmtheader.ddpfPixelFormat.dwFourCC == (('D'<<0)|('X'<<8)|('T'<<16)|('3'<<24)))
	{
//		if (tex->flags & IF_PREMULTIPLYALPHA)
//			return false;
		encoding = PTI_BC2_RGBA;
//		pad = 8;
	}
	else if (*(int*)&fmtheader.ddpfPixelFormat.dwFourCC == (('D'<<0)|('X'<<8)|('T'<<16)|('4'<<24)))	//dx5 with premultiplied alpha
	{
//		if (!(tex->flags & IF_PREMULTIPLYALPHA))
//			return false;

		encoding = PTI_BC3_RGBA;
//		pad = 8;
	}
	else if (*(int*)&fmtheader.ddpfPixelFormat.dwFourCC == (('D'<<0)|('X'<<8)|('T'<<16)|('5'<<24)))
	{
//		if (tex->flags & IF_PREMULTIPLYALPHA)
//			return false;

		encoding = PTI_BC3_RGBA;
//		pad = 8;
	}
	else if (*(int*)&fmtheader.ddpfPixelFormat.dwFourCC == (('E'<<0)|('T'<<8)|('C'<<16)|('2'<<24)))
	{
		encoding = PTI_ETC2_RGB8;
//		pad = 8;
	}
	else if (*(int*)&fmtheader.ddpfPixelFormat.dwFourCC == (('D'<<0)|('X'<<8)|('1'<<16)|('0'<<24)))
	{
		//this has some weird extra header with dxgi format types.
		memcpy(&fmt10header, filedata+4+fmtheader.dwSize, sizeof(fmt10header));
		fmtheader.dwSize += sizeof(fmt10header);
//		pad = 8;
		switch(fmt10header.dxgiformat)
		{
		case 2/*DXGI_FORMAT_R32G32B32A32_FLOAT*/:
			encoding = PTI_RGBA32F;
			break;
		case 10/*DXGI_FORMAT_R16G16B16A16_FLOAT*/:
			encoding = PTI_RGBA16F;
			break;
		case 24/*DXGI_FORMAT_R10G10B10A2_UNORM*/:
			encoding = PTI_A2BGR10;
			break;
		case 28/*DXGI_FORMAT_R8G8B8A8_UNORM*/:
			encoding = PTI_RGBA8;
			break;
		case 67/*DXGI_FORMAT_R9G9B9E5_SHAREDEXP*/:
			encoding = PTI_E5BGR9;
			break;
		case 71/*DXGI_FORMAT_BC1_UNORM*/:
			encoding = PTI_BC1_RGBA;
			break;
		case 72/*DXGI_FORMAT_BC1_UNORM_SRGB*/:
			encoding = PTI_BC1_RGBA_SRGB;
			break;
		case 74/*DXGI_FORMAT_BC2_UNORM*/:
			encoding = PTI_BC2_RGBA;
			break;
		case 75/*DXGI_FORMAT_BC2_UNORM_SRGB*/:
			encoding = PTI_BC2_RGBA_SRGB;
			break;
		case 77/*DXGI_FORMAT_BC3_UNORM*/:
			encoding = PTI_BC3_RGBA;
			break;
		case 78/*DXGI_FORMAT_BC3_UNORM_SRGB*/:
			encoding = PTI_BC3_RGBA_SRGB;
			break;
		case 80/*DXGI_FORMAT_BC4_UNORM*/:
			encoding = PTI_BC4_R8;
			break;
		case 81/*DXGI_FORMAT_BC4_SNORM*/:
			encoding = PTI_BC4_R8_SNORM;
			break;
		case 83/*DXGI_FORMAT_BC5_UNORM*/:
			encoding = PTI_BC5_RG8;
			break;
		case 84/*DXGI_FORMAT_BC5_SNORM*/:
			encoding = PTI_BC5_RG8_SNORM;
			break;
		case 85/*DXGI_FORMAT_B5G6R5_UNORM*/:
			encoding = PTI_RGB565;
			break;
		case 86/*DXGI_FORMAT_B5G5R5A1_UNORM*/:
			encoding = PTI_ARGB1555;
			break;
		case 87/*DXGI_FORMAT_B8G8R8A8_UNORM*/:
			encoding = PTI_BGRA8;
			break;
		case 95/*DXGI_FORMAT_BC6H_UF16*/:
			encoding = PTI_BC6_RGB_UFLOAT;
			break;
		case 96/*DXGI_FORMAT_BC6H_SF16*/:
			encoding = PTI_BC6_RGB_SFLOAT;
			break;
		case 98/*DXGI_FORMAT_BC7_UNORM*/:
			encoding = PTI_BC7_RGBA;
			break;
		case 99/*DXGI_FORMAT_BC7_UNORM_SRGB*/:
			encoding = PTI_BC7_RGBA_SRGB;
			break;

		default:
			Con_Printf("Unsupported dds10 dxgi in %s - %u\n", fname, fmt10header.dxgiformat);
			return NULL;
		}
	}
	else
	{
		Con_Printf("Unsupported dds fourcc in %s - \"%c%c%c%c\"\n", fname,
			((char*)&fmtheader.ddpfPixelFormat.dwFourCC)[0],
			((char*)&fmtheader.ddpfPixelFormat.dwFourCC)[1],
			((char*)&fmtheader.ddpfPixelFormat.dwFourCC)[2],
			((char*)&fmtheader.ddpfPixelFormat.dwFourCC)[3]);
		return NULL;
	}

	if ((fmtheader.ddsCaps[1] & 0x200) && (fmtheader.ddsCaps[1] & 0xfc00) != 0xfc00)
		return NULL;	//cubemap without all 6 faces defined.
	if (fmtheader.ddsCaps[1] & 0x200000)
		return NULL;	//3d texture. fte internally interleaves layers on the x axis. I'll bet dds does not.

	Image_BlockSizeForEncoding(encoding, &blockbytes, &blockwidth, &blockheight);
	if (!blockbytes)
		return NULL;	//werid/unsupported

	mips = Z_Malloc(sizeof(*mips));
	mips->mipcount = 0;
	if (fmtheader.ddsCaps[1] & 0x200)
	{
		layers = 6;
		mips->type = PTI_CUBEMAP;
	}
	else if (fmtheader.ddsCaps[1] & 0x200000)
		mips->type = PTI_3D;
	else
		mips->type = PTI_2D;
	mips->extrafree = filedata;
	mips->encoding = encoding;

	filedata += 4+fmtheader.dwSize;

	datasize = fmtheader.dwPitchOrLinearSize;
	w = fmtheader.dwWidth;
	h = fmtheader.dwHeight;
	for (mipnum = 0; mipnum < nummips; mipnum++)
	{
		if (mips->mipcount >= countof(mips->mip))
			break;

//		if (datasize < 8)
//			datasize = pad;
		datasize = ((w+blockwidth-1)/blockwidth) * ((h+blockheight-1)/blockheight) * blockbytes;

		for (layer = 0; layer < layers; layer++)
		{
			mips->mip[mips->mipcount].data = filedata;
			mips->mip[mips->mipcount].datasize = datasize;
			mips->mip[mips->mipcount].width = w;
			mips->mip[mips->mipcount].height = h;
			mips->mip[mips->mipcount].depth = 1;
			mips->mipcount++;
			filedata += datasize;
		}
		w = (w+1)>>1;
		h = (h+1)>>1;
	}

	return mips;
}

qboolean Image_WriteDDSFile(const char *filename, enum fs_relative fsroot, struct pendingtextureinfo *mips)
{
	vfsfile_t *file;
	size_t mipnum;
	size_t a;
	dds10header_t h10={0};
	ddsheader h9={0};

	unsigned int blockbytes, blockwidth, blockheight;
	unsigned int arraysize = (mips->type==PTI_CUBEMAP||mips->type==PTI_CUBEMAP_ARRAY)?6:1;

	Image_BlockSizeForEncoding(mips->encoding, &blockbytes, &blockwidth, &blockheight);

	h9.dwSize = sizeof(h9);
	h9.dwFlags = 0;
	h9.dwFlags |= 1;			//CAPS
	h9.dwFlags |= 2;			//HEIGHT
	h9.dwFlags |= 4;			//WIDTH
	h9.dwFlags |= 0x1000;		//PIXELFORMAT
	if (blockwidth != 1 || blockheight != 1)
	{
		h9.dwFlags |= 0x80000;	//LINEARSIZE
		h9.dwPitchOrLinearSize = ((mips->mip[0].width+blockwidth-1)/blockwidth)*((mips->mip[0].height+blockheight-1)/blockheight)*blockbytes;
	}
	else
	{
		h9.dwFlags |= 8;		//PITCH
		h9.dwPitchOrLinearSize = mips->mip[0].width*blockbytes;
	}
	if (mips->mipcount > 1)
		h9.dwFlags |= 0x20000;	//MIPMAPCOUNT
	h9.dwHeight = mips->mip[0].height;
	h9.dwWidth = mips->mip[0].width;
	h9.dwDepth = 0;
	h9.dwMipMapCount = mips->mipcount/arraysize;
	h9.ddpfPixelFormat.dwSize = 32;
	h9.ddpfPixelFormat.dwFlags = 4/*DDPF_FOURCC*/;
	h9.ddpfPixelFormat.dwFourCC = ('D'<<0)|('X'<<8)|('1'<<16)|('0'<<24);
	h9.ddsCaps[0] = 0x1000;		//TEXTURE
	if (mips->mipcount > 1)
		h9.ddsCaps[0] |= 0x8;		//COMPLEX
	if (mips->mipcount > arraysize)
		h9.ddsCaps[0] |= 0x400000;	//MIPMAP
	h9.ddsCaps[1] = 0;
	h10.miscflag = 0;
	h10.arraysize = arraysize;
	h10.miscflags2 = 0;

	switch(mips->type)
	{
	case PTI_3D:
		h9.ddsCaps[1] |= 0x200000;	//VOLUME
		h10.resourcetype = 4;	//3d
		break;
	case PTI_CUBEMAP:
	case PTI_CUBEMAP_ARRAY:
		h9.ddsCaps[1] |= 0x200|0xfc00;		//CUBEMAP+faces
		h10.resourcetype = 3;	//2d
		h10.miscflag = 4;//DDS_RESOURCE_MISC_TEXTURECUBE - otherwise they're basicaly just 2d_arrays
		break;
	case PTI_2D:
	case PTI_2D_ARRAY:
		h10.resourcetype = 3;	//2d
		break;
	}

	h10.dxgiformat = 0;

	switch(mips->encoding)
	{
	case PTI_ETC1_RGB8:
	case PTI_ETC2_RGB8:
	case PTI_ETC2_RGB8_SRGB:
	case PTI_ETC2_RGB8A1:
	case PTI_ETC2_RGB8A1_SRGB:
	case PTI_ETC2_RGB8A8:
	case PTI_ETC2_RGB8A8_SRGB:
	case PTI_EAC_R11:
	case PTI_EAC_R11_SNORM:
	case PTI_EAC_RG11:
	case PTI_EAC_RG11_SNORM:
	case PTI_ASTC_4X4:
	case PTI_ASTC_5X4:
	case PTI_ASTC_5X5:
	case PTI_ASTC_6X5:
	case PTI_ASTC_6X6:
	case PTI_ASTC_8X5:
	case PTI_ASTC_8X6:
	case PTI_ASTC_10X5:
	case PTI_ASTC_10X6:
	case PTI_ASTC_8X8:
	case PTI_ASTC_10X8:
	case PTI_ASTC_10X10:
	case PTI_ASTC_12X10:
	case PTI_ASTC_12X12:
	case PTI_ASTC_4X4_SRGB:
	case PTI_ASTC_5X4_SRGB:
	case PTI_ASTC_5X5_SRGB:
	case PTI_ASTC_6X5_SRGB:
	case PTI_ASTC_6X6_SRGB:
	case PTI_ASTC_8X5_SRGB:
	case PTI_ASTC_8X6_SRGB:
	case PTI_ASTC_10X5_SRGB:
	case PTI_ASTC_10X6_SRGB:
	case PTI_ASTC_8X8_SRGB:
	case PTI_ASTC_10X8_SRGB:
	case PTI_ASTC_10X10_SRGB:
	case PTI_ASTC_12X10_SRGB:
	case PTI_ASTC_12X12_SRGB:	return false;	//unsupported
	case PTI_BC1_RGB:
	case PTI_BC1_RGBA:			h10.dxgiformat = 71/*DXGI_FORMAT_BC1_UNORM*/; break;
	case PTI_BC1_RGB_SRGB:
	case PTI_BC1_RGBA_SRGB:		h10.dxgiformat = 72/*DXGI_FORMAT_BC1_UNORM_SRGB*/; break;
	case PTI_BC2_RGBA:			h10.dxgiformat = 74/*DXGI_FORMAT_BC2_UNORM*/; break;
	case PTI_BC2_RGBA_SRGB:		h10.dxgiformat = 75/*DXGI_FORMAT_BC2_UNORM_SRGB*/; break;
	case PTI_BC3_RGBA:			h10.dxgiformat = 77/*DXGI_FORMAT_BC3_UNORM*/; break;
	case PTI_BC3_RGBA_SRGB:		h10.dxgiformat = 78/*DXGI_FORMAT_BC3_UNORM_SRGB*/; break;
	case PTI_BC4_R8_SNORM:		h10.dxgiformat = 81/*DXGI_FORMAT_BC4_SNORM*/; break;
	case PTI_BC4_R8:			h10.dxgiformat = 80/*DXGI_FORMAT_BC4_UNORM*/; break;
	case PTI_BC5_RG8_SNORM:		h10.dxgiformat = 84/*DXGI_FORMAT_BC5_SNORM*/; break;
	case PTI_BC5_RG8:			h10.dxgiformat = 83/*DXGI_FORMAT_BC5_UNORM*/; break;
	case PTI_BC6_RGB_UFLOAT:	h10.dxgiformat = 95/*DXGI_FORMAT_BC6H_UF16*/; break;
	case PTI_BC6_RGB_SFLOAT:	h10.dxgiformat = 96/*DXGI_FORMAT_BC6H_SF16*/; break;
	case PTI_BC7_RGBA:			h10.dxgiformat = 98/*DXGI_FORMAT_BC7_UNORM*/; break;
	case PTI_BC7_RGBA_SRGB:		h10.dxgiformat = 99/*DXGI_FORMAT_BC7_UNORM_SRGB*/; break;

	case PTI_BGRA8:				h10.dxgiformat = 87/*DXGI_FORMAT_B8G8R8A8_UNORM*/; break;
	case PTI_RGBA8:				h10.dxgiformat = 28/*DXGI_FORMAT_R8G8B8A8_UNORM*/; break;
	case PTI_BGRA8_SRGB:		h10.dxgiformat = 91/*DXGI_FORMAT_B8G8R8A8_UNORM_SRGB*/; break;
	case PTI_RGBA8_SRGB:		h10.dxgiformat = 29/*DXGI_FORMAT_R8G8B8A8_UNORM_SRGB*/; break;
	case PTI_L8:				return false;	//unsupported
	case PTI_L8A8:				return false;	//unsupported
	case PTI_L8_SRGB:			return false;	//unsupported
	case PTI_L8A8_SRGB:			return false;	//unsupported
	case PTI_RGB8:				return false;	//unsupported
	case PTI_BGR8:				return false;	//unsupported
	case PTI_R16:				h10.dxgiformat = 56/*DXGI_FORMAT_R16_UNORM*/; break;
	case PTI_RGBA16:			h10.dxgiformat = 11/*DXGI_FORMAT_R16G16B16A16_UNORM*/; break;
	case PTI_R16F:				h10.dxgiformat = 54/*DXGI_FORMAT_R16_FLOAT*/; break;
	case PTI_R32F:				h10.dxgiformat = 41/*DXGI_FORMAT_R32_FLOAT*/; break;
	case PTI_RGBA16F:			h10.dxgiformat = 10/*DXGI_FORMAT_R16G16B16A16_FLOAT*/; break;
	case PTI_RGBA32F:			h10.dxgiformat = 2/*DXGI_FORMAT_R32G32B32A32_FLOAT*/; break;
	case PTI_A2BGR10:			h10.dxgiformat = 24/*DXGI_FORMAT_R10G10B10A2_UNORM*/; break;
	case PTI_E5BGR9:			h10.dxgiformat = 67/*DXGI_FORMAT_R9G9B9E5_SHAREDEXP*/; break;
	case PTI_P8:
	case PTI_R8:				h10.dxgiformat = 61/*DXGI_FORMAT_R8_UNORM*/; break;
	case PTI_RG8:				h10.dxgiformat = 49/*DXGI_FORMAT_R8G8_UNORM*/; break;
	case PTI_R8_SNORM:			h10.dxgiformat = 63/*DXGI_FORMAT_R8_SNORM*/; break;
	case PTI_RG8_SNORM:			h10.dxgiformat = 51/*DXGI_FORMAT_R8G8_SNORM*/; break;
	case PTI_BGRX8:				h10.dxgiformat = 88/*DXGI_FORMAT_B8G8R8X8_UNORM*/; break;
	case PTI_RGBX8:				return false;	//unsupported
	case PTI_BGRX8_SRGB:		h10.dxgiformat = 93/*DXGI_FORMAT_B8G8R8X8_UNORM_SRGB*/; break;
	case PTI_RGBX8_SRGB:		return false; //unsupported
	case PTI_RGB565:			h10.dxgiformat = 85/*DXGI_FORMAT_B5G6R5_UNORM*/; break;
	case PTI_RGBA4444:			return false;	//unsupported
	case PTI_ARGB4444:			h10.dxgiformat = 115/*DXGI_FORMAT_B4G4R4A4_UNORM*/; break;
	case PTI_RGBA5551:			return false;	//unsupported
	case PTI_ARGB1555:			h10.dxgiformat = 86/*DXGI_FORMAT_B5G5R5A1_UNORM*/; break;
	case PTI_DEPTH16:			h10.dxgiformat = 55/*DXGI_FORMAT_D16_UNORM*/; break;
	case PTI_DEPTH24:			return false; //unsupported
	case PTI_DEPTH32:			h10.dxgiformat = 40/*DXGI_FORMAT_D32_FLOAT*/; break;
	case PTI_DEPTH24_8:			h10.dxgiformat = 45/*DXGI_FORMAT_D24_UNORM_S8_UINT*/; break;

#ifdef FTE_TARGET_WEB
	case PTI_WHOLEFILE:
#endif
	case PTI_EMULATED:
	case PTI_MAX:
		return false;

//	default:
//		return;
	}

	//truncate the mip chain if they're dodgy sizes.
	for (mipnum = 1; mipnum < h9.dwMipMapCount; mipnum++)
	{
		size_t m = mipnum*arraysize;
		size_t p = (mipnum-1)*arraysize;
		if (mips->mip[m].width != max(1,(mips->mip[p].width)>>1) ||
			mips->mip[m].height != max(1,(mips->mip[p].height)>>1))
		{
			h9.dwMipMapCount = mipnum;
			break;
		}
	}

	if (strchr(filename, '*') || strchr(filename, ':'))
		return false;

	file = FS_OpenVFS(filename, "wb", FS_GAMEONLY);
	if (!file)
		return false;
	VFS_WRITE(file, "DDS ", 4);
	VFS_WRITE(file, &h9, sizeof(h9));
	VFS_WRITE(file, &h10, sizeof(h10));

	//our internal state uses a0m0, a1m0, a0m1, a1m1
	//DDS requires a0m0, a0m1, a1m0, a1m1, so reorder with two nested loops
	for (a = 0; a < arraysize; a++)
	{
		for (mipnum = 0; mipnum < h9.dwMipMapCount; mipnum++)
		{
			size_t m = a + mipnum*arraysize;
			VFS_WRITE(file, mips->mip[m].data, mips->mip[m].datasize);
		}
	}

	VFS_CLOSE(file);
	return true;
}
#endif

#ifdef IMAGEFMT_BLP
static struct pendingtextureinfo *Image_ReadBLPFile(unsigned int flags, const char *fname, qbyte *filedata, size_t filesize)
{
	//FIXME: cba with endian.
	int miplevel;
	int w, h, i;
	struct blp_s
	{
		char blp2[4];
		int type;
		qbyte encoding;
		qbyte alphadepth;
		qbyte alphaencoding;
		qbyte hasmips;
		unsigned int xres;
		unsigned int yres;
		unsigned int mipoffset[16];
		unsigned int mipsize[16];
		unsigned int palette[256];
	} *blp;
	unsigned int *tmpmem = NULL;
	unsigned char *in;
	unsigned int inlen;

	struct pendingtextureinfo *mips;

	blp = (void*)filedata;

	if (memcmp(blp->blp2, "BLP2", 4) || blp->type != 1)
		return NULL;

	mips = Z_Malloc(sizeof(*mips));
	mips->mipcount = 0;
	mips->type = PTI_2D;

	w = LittleLong(blp->xres);
	h = LittleLong(blp->yres);

	if (blp->encoding == 2)
	{
		//s3tc/dxt
		switch(blp->alphaencoding)
		{
		default:
		case 0: //dxt1
			if (blp->alphadepth)
				mips->encoding = PTI_BC1_RGBA;
			else
				mips->encoding = PTI_BC1_RGB;
			break;
		case 1: //dxt2/3
			mips->encoding = PTI_BC2_RGBA;
			break;
		case 7: //dxt4/5
			mips->encoding = PTI_BC3_RGBA;
			break;
		}
		for (miplevel = 0; miplevel < 16; )
		{
			if (!w && !h)	//shrunk to no size
				break;
			if (!w)
				w = 1;
			if (!h)
				h = 1;
			if (!blp->mipoffset[miplevel] || !blp->mipsize[miplevel] || blp->mipoffset[miplevel]+blp->mipsize[miplevel] > filesize)	//no data
				break;
			mips->mip[miplevel].width = w;
			mips->mip[miplevel].height = h;
			mips->mip[miplevel].depth = 1;
			mips->mip[miplevel].data = filedata + LittleLong(blp->mipoffset[miplevel]);
			mips->mip[miplevel].datasize = LittleLong(blp->mipsize[miplevel]);

			miplevel++;
			if (!blp->hasmips || (flags & IF_NOMIPMAP))
				break;
			w >>= 1;
			h >>= 1;
		}
		mips->mipcount = miplevel;
		mips->extrafree = filedata;
	}
	else
	{
		mips->encoding = PTI_BGRA8;
		for (miplevel = 0; miplevel < 16; )
		{
			if (!w && !h)
				break;
			if (!w)
				w = 1;
			if (!h)
				h = 1;
			//if we ran out of mips to load, give up.
			if (!blp->mipoffset[miplevel] || !blp->mipsize[miplevel] || blp->mipoffset[miplevel]+blp->mipsize[miplevel] > filesize)
			{
				//if we got at least one mip, cap the mips. might help save some ram? naaah...
				//if this is the first mip, well, its completely fucked.
				break;
			}

			in = filedata + LittleLong(blp->mipoffset[miplevel]);
			inlen = LittleLong(blp->mipsize[miplevel]);

			if (inlen != w*h+((w*h*blp->alphadepth+7)>>3))
			{
				Con_Printf("%s: mip level %i does not contain the correct amount of data\n", fname, miplevel);
				break;
			}

			mips->mip[miplevel].width = w;
			mips->mip[miplevel].height = h;
			mips->mip[miplevel].depth = 1;
			mips->mip[miplevel].datasize = 4*w*h;
			mips->mip[miplevel].data = tmpmem = BZ_Malloc(4*w*h);
			mips->mip[miplevel].needfree = true;

			//load the rgb data first (8-bit paletted)
			for (i = 0; i < w*h; i++)
				tmpmem[i] = blp->palette[*in++] | 0xff000000;

			//and then change the alpha bits accordingly.
			switch(blp->alphadepth)
			{
			case 0:
				//BGRX palette, 8bit
				break;
			case 1:
				//BGRX palette, 8bit
				//1bit trailing alpha
				for (i = 0; i < w*h; i+=8, in++)
				{
					tmpmem[i+0] = (tmpmem[i+0] & 0xffffff) | ((*in&0x01)?0xff000000:0);
					tmpmem[i+1] = (tmpmem[i+1] & 0xffffff) | ((*in&0x02)?0xff000000:0);
					tmpmem[i+2] = (tmpmem[i+2] & 0xffffff) | ((*in&0x04)?0xff000000:0);
					tmpmem[i+3] = (tmpmem[i+3] & 0xffffff) | ((*in&0x08)?0xff000000:0);
					tmpmem[i+4] = (tmpmem[i+4] & 0xffffff) | ((*in&0x10)?0xff000000:0);
					tmpmem[i+5] = (tmpmem[i+5] & 0xffffff) | ((*in&0x20)?0xff000000:0);
					tmpmem[i+6] = (tmpmem[i+6] & 0xffffff) | ((*in&0x40)?0xff000000:0);
					tmpmem[i+7] = (tmpmem[i+7] & 0xffffff) | ((*in&0x80)?0xff000000:0);
				}
				break;
			case 4:
				//BGRX palette, 8bit
				//4bit trailing alpha
				for (i = 0; i < w*h; i++)
					tmpmem[i] = (tmpmem[i] & 0xffffff) | (*in++*0x11000000);
				break;
			case 8:
				//BGRX palette, 8bit
				//8bit trailing alpha
				for (i = 0; i < w*h; i++)
					tmpmem[i] = (tmpmem[i] & 0xffffff) | (*in++<<24);
				break;
			}

			miplevel++;
			if (!blp->hasmips || (flags & IF_NOMIPMAP))
				break;
			w = w>>1;
			h = h>>1;
		}
		BZ_Free(filedata);
		mips->mipcount = miplevel;
	}
	return mips;
}
#endif

#ifdef IMAGEFMT_VTF
//many of these look like dupes, not really sure how they're meant to work. probably legacy.
typedef enum {
	VMF_INVALID=-1,
//	VMF_RGBA8=0,
//	VMF_ABGR8=1,
	VMF_RGB8=2,
	VMF_BGR8=3,
//	VMF_RGB565=4,
//	VMF_I8=5,
//	VMF_IA8=6,
//	VMF_P8=7,
//	VMF_A8=8,
//	VMF_RGB8_BS=9,
//	VMF_BGR8_BS=10,
//	VMF_ARGB_BS=11,
	VMF_BGRA8=12,
	VMF_BC1=13,
	VMF_BC2=14,
	VMF_BC3=15,
	VMF_BGRX8=16,
//	VMF_BGR565=17,
//	VMF_BGRX5551=18,
//	VMF_BGRA4444=19,
	VMF_BC1A=20,
//	VMF_BGRA5551=21,
	VMF_UV88=22,
//	VMF_UVWQ8=23,
	VMF_RGBA16F=24,
//	VMF_RGBA16N=25,
//	VMF_UVLX8=26,
	VMF_MAX
} fmtfmt_t;
static uploadfmt_t ImageVTF_VtfToFTE(fmtfmt_t f)
{
	switch(f)
	{
	case VMF_BC1:
		return PTI_BC1_RGB;
	case VMF_BC1A:
		return PTI_BC1_RGBA;
	case VMF_BC2:
		return PTI_BC2_RGBA;
	case VMF_BC3:
		return PTI_BC3_RGBA;
	case VMF_RGB8:
		return PTI_RGB8;
	case VMF_BGR8:
		return PTI_BGR8;
	case VMF_BGRA8:
		return PTI_BGRA8;
	case VMF_BGRX8:
		return PTI_BGRX8;
	case VMF_RGBA16F:
		return PTI_RGBA16F;
	case VMF_UV88:
		return PTI_RG8;
	case VMF_INVALID:
		return PTI_INVALID;

	default:
		return PTI_INVALID;
	}
}
static struct pendingtextureinfo *Image_ReadVTFFile(unsigned int flags, const char *fname, qbyte *filedata, size_t filesize)
{
	//FIXME: cba with endian.
	struct vtf_s
	{
		char magic[4];
		unsigned int major,minor;
		unsigned int headersize;

		unsigned short width, height;
		unsigned int flags;
		unsigned short numframes, firstframe;
		unsigned int pad1;

		vec3_t reflectivity;
		float pad2;

		float bumpmapscale;
		unsigned int imgformat;
		unsigned char mipmapcount;
		unsigned char lowresfmt_misaligned[4];
		unsigned char lowreswidth;
		unsigned char lowresheight;

		//7.2
		unsigned char depth_misaligned[2];
		//7.3
		unsigned char pad3[3];
		unsigned int numresources;
	} *vtf;
	fmtfmt_t vmffmt, lrfmt;
	unsigned int bw, bh, bb;
	qbyte *end = filedata + filesize;
	unsigned int face, faces, frame, frames, miplevel, miplevels, img;
	unsigned int w, h;
	size_t	datasize;
	unsigned int version;

	struct pendingtextureinfo *mips;

	vtf = (void*)filedata;

	if (memcmp(vtf->magic, "VTF\0", 4))
		return NULL;

	version = (vtf->major<<16)|vtf->minor;
	if (version >= 0x00070003)
		return NULL;	//we don't support the whole resources thing.

	lrfmt = (vtf->lowresfmt_misaligned[0]<<0)|(vtf->lowresfmt_misaligned[1]<<16)|(vtf->lowresfmt_misaligned[2]<<16)|(vtf->lowresfmt_misaligned[3]<<24);
	vmffmt = vtf->imgformat;

	if (vtf->lowreswidth && vtf->lowresheight)
		Image_BlockSizeForEncoding(ImageVTF_VtfToFTE(lrfmt), &bb, &bw, &bh);
	else
		bb=bw=bh=1;
	datasize = ((vtf->lowreswidth+bw-1)/bw) * ((vtf->lowresheight+bh-1)/bh) * bb;

	mips = Z_Malloc(sizeof(*mips));
	mips->type = (vtf->flags & 0x4000)?PTI_CUBEMAP:PTI_2D;

	mips->extrafree = filedata;
	filedata += vtf->headersize;	//skip the header
	filedata += datasize;			//and skip the low-res image too.

	mips->encoding = ImageVTF_VtfToFTE(vmffmt);
	Image_BlockSizeForEncoding(mips->encoding, &bb, &bw, &bh);

	miplevels = vtf->mipmapcount;
	frames = 1;//vtf->numframes;
	faces = ((mips->type==PTI_CUBEMAP)?6:1);	//no cubemaps yet.

	mips->mipcount = miplevels * frames * faces;
	while (mips->mipcount > countof(mips->mip))
	{
		if (miplevels > 1)
			miplevels--;
		else
			frames--;
		mips->mipcount = miplevels * frames * faces;
	}
	if (!mips->mipcount)
	{
		Z_Free(mips);
		return NULL;
	}
	for (miplevel = vtf->mipmapcount; miplevel-- > 0;)
	{	//smallest to largest, which is awkward.
		w = vtf->width>>miplevel;
		h = vtf->height>>miplevel;
		if (!w)
			w = 1;
		if (!h)
			h = 1;
		datasize = ((w+bw-1)/bw) * ((h+bh-1)/bh) * bb;
		for (frame = 0; frame < vtf->numframes; frame++)
		{
			for (face = 0; face < faces; face++)
			{
				if (miplevel < miplevels && face < faces)
				{
					img = face+miplevel*faces + frame*miplevels*faces;
					if (img >= countof(mips->mip))
						break;	//erk?
					if (filedata + datasize > end)
						break;	//no more data here...
					mips->mip[img].width = w;
					mips->mip[img].height = h;
					mips->mip[img].depth = 1;
					mips->mip[img].data = filedata;
					mips->mip[img].datasize = datasize;
				}
				filedata += datasize;
			}
		}
	}
	return mips;
}
#endif

//if force_rgba8 then it guarentees rgba8 or rgbx8, otherwise can return l8, etc
qbyte *ReadRawImageFile(qbyte *buf, int len, int *width, int *height, uploadfmt_t *format, qboolean force_rgba8, const char *fname)
{
	qbyte *data;
	*format = PTI_RGBX8;
	if ((data = ReadTargaFile(buf, len, width, height, format, false, force_rgba8?PTI_RGBA8:PTI_INVALID)))
	{
		TRACE(("dbg: Read32BitImageFile: tga\n"));
		return data;
	}

#ifdef AVAIL_PNGLIB
	if (len > 4 && (buf[0] == 137 && buf[1] == 'P' && buf[2] == 'N' && buf[3] == 'G') && (data = ReadPNGFile(fname, buf, len, width, height, format)))
	{
		TRACE(("dbg: Read32BitImageFile: png\n"));
		return data;
	}
#endif
#ifdef AVAIL_JPEGLIB
	//jpeg jfif only.
	if (len > 4 && (buf[0] == 0xff && buf[1] == 0xd8 && buf[2] == 0xff /*&& buf[3] == 0xe0*/) && (data = ReadJPEGFile(buf, len, width, height)))
	{
		*format = PTI_RGBX8;
		TRACE(("dbg: Read32BitImageFile: jpeg\n"));
		return data;
	}
#endif
#ifdef IMAGEFMT_PCX
	if ((data = ReadPCXFile(buf, len, width, height)))
	{
		*format = PTI_RGBA8;
		TRACE(("dbg: Read32BitImageFile: pcx\n"));
		return data;
	}
#endif

#ifdef IMAGEFMT_BMP
	if (len > 2 && (buf[0] == 'B' && buf[1] == 'M') && (data = ReadBMPFile(buf, len, width, height)))
	{
		*format = PTI_RGBA8;
		TRACE(("dbg: Read32BitImageFile: bmp\n"));
		return data;
	}

	if (len > 6 && buf[0]==0&&buf[1]==0 && buf[2]==1&&buf[3]==0 && (data = ReadICOFile(fname, buf, len, width, height, format)))
	{
		TRACE(("dbg: Read32BitImageFile: ico\n"));
		return data;
	}
#endif

#ifdef IMAGEFMT_PBM
	if (len > 2 && buf[0] == 'P' && ((buf[1] >= '1' && buf[1] <= '7') || buf[1] == 'F' || buf[1] == 'f') && (data = ReadPBMFile(buf, len, fname, width, height, format)))
		return data;
#endif
#ifdef IMAGEFMT_HDR
	if (len > 10 && (!strncmp(buf, "#?RADIANCE", 10)||!strncmp(buf, "#?RGBE", 6)) && (data = ReadRadianceFile(buf, len, fname, width, height, format)))
		return data;
#endif
#ifdef IMAGEFMT_PSD
	if (len > 26 && !strncmp(buf, "8BPS", 4) && (data = ReadPSDFile(buf, len, fname, width, height, format)))
		return data;
#endif

#if 1//def IMAGEFMT_LMP
	if (len >= 8)	//.lmp has no magic id. guess at it.
	{
		int w = LittleLong(((int*)buf)[0]);
		int h = LittleLong(((int*)buf)[1]);
		int i;
		if (w >= 3 && h	>= 4 && w*h+sizeof(int)*2 == len)
		{	//quake lmp
			qboolean foundalpha = false;
			qbyte *in = buf+sizeof(int)*2;
			data = BZ_Malloc(w * h * sizeof(int));
			for (i = 0; i < w * h; i++)
			{
				if (in[i] == 255)
					foundalpha = true;
				((unsigned int*)data)[i] = d_8to24rgbtable[in[i]];
			}
			*width = w;
			*height = h;
			*format = foundalpha?PTI_RGBA8:PTI_RGBX8;
			return data;
		}
		else if (w >= 3 && h >= 4 && w*h+sizeof(int)*2+768+2 == len)
		{	//halflife. should probably verify that those 2 extra bytes read as 256.
			qboolean foundalpha = false;
			qbyte *in = buf+sizeof(int)*2;
			qbyte *palette = in + w*h+2, *p;
			data = BZ_Malloc(w * h * sizeof(int));
			for (i = 0; i < w * h; i++)
			{
				if (in[i] == 255)
					foundalpha = true;
				p = palette + 3*in[i];
				data[(i<<2)+0] = p[2];
				data[(i<<2)+1] = p[1];
				data[(i<<2)+2] = p[0];
				data[(i<<2)+3] = 255;
			}
			*width = w;
			*height = h;
			*format = foundalpha?PTI_RGBA8:PTI_RGBX8;
			return data;
		}
	}
#endif

	TRACE(("dbg: Read32BitImageFile: life sucks\n"));

	return NULL;
}

static void *R_FlipImage32(void *in, int *inoutwidth, int *inoutheight, qboolean flipx, qboolean flipy, qboolean flipd)
{
	int x, y;
	unsigned int *in32, *inr, *out32;
	void *out;
	int inwidth = *inoutwidth;
	int inheight = *inoutheight;
	int rowstride = inwidth;
	int colstride = 1;

	//simply return if no operation
	if (!flipx && !flipy && !flipd)
		return in;

	inr = in;
	out32 = out = BZ_Malloc(inwidth*inheight*4);

	if (flipy)
	{
		inr += inwidth*inheight-inwidth;//start on the bottom row
		rowstride *= -1;	//and we need to move up instead
	}
	if (flipx)
	{
		colstride *= -1;	//move backwards
		inr += inwidth-1;	//start at the end of the row
	}
	if (flipd)
	{ 
		//switch the dimensions
		int tmp = inwidth;
		inwidth = inheight;
		inheight = tmp;
		//make sure the caller gets the new dimensions
		*inoutwidth = inwidth;
		*inoutheight = inheight;
		//switch the strides
		tmp = colstride;
		colstride = rowstride;
		rowstride = tmp;
	}

	//rows->rows, columns->columns
	for (y = 0; y < inheight; y++)
	{
		in32 = inr;	//reset the input after each row, so we have truely independant row+column strides
		inr += rowstride;
		for (x = 0; x < inheight; x++)
		{
			*out32++ = *in32;
			in32 += colstride;
		}
	}
	BZ_Free(in);
	return out;
}

static int tex_extensions_count;
#define tex_extensions_max 15
static struct
{
	char name[6];
} tex_extensions[tex_extensions_max];
static void QDECL R_ImageExtensions_Callback(struct cvar_s *var, char *oldvalue)
{
	char *v = var->string;
	tex_extensions_count = 0;

	while (tex_extensions_count < tex_extensions_max)
	{
		v = COM_Parse(v);
		if (!v)
			break;
		Q_snprintfz(tex_extensions[tex_extensions_count].name, sizeof(tex_extensions[tex_extensions_count].name), ".%s", com_token); 
		tex_extensions_count++;
	}

	if (tex_extensions_count < tex_extensions_max)
	{
		Q_snprintfz(tex_extensions[tex_extensions_count].name, sizeof(tex_extensions[tex_extensions_count].name), ""); 
		tex_extensions_count++;
	}
}

static struct
{
	char *path;
	int args;

	int enabled;
} tex_path[] =
{
	/*if three args, first is the subpath*/
	/*the last two args are texturename then extension*/
	{"%s%s",			2, 1},	/*directly named texture*/
	{"textures/%s/%s%s",3, 1},	/*fuhquake compatibility*/
	{"%s/%s%s",			3, 1},	/*fuhquake compatibility*/
	{"textures/%s%s",	2, 1},	/*directly named texture with textures/ prefix*/
#ifdef HAVE_LEGACY
	{"override/%s%s",	2, 1}	/*tenebrae compatibility*/
#endif
};

static void Image_MipMap4X8 (qbyte *in, int inwidth, int inheight, qbyte *out, int outwidth, int outheight)
{
	int		i, j;
	qbyte	*inrow;

	int rowwidth = inwidth*4;	//rowwidth is the byte width of the input
	inrow = in;

	//mips round down, except for when the input is 1. which bugs out.
	if (inwidth <= 1 && inheight <= 1)
	{
		out[0] = in[0];
		out[1] = in[1];
		out[2] = in[2];
		out[3] = in[3];
	}
	else if (inheight <= 1)
	{
		//single row, don't peek at the next
		for (in = inrow, j=0 ; j<outwidth ; j++, out+=4, in+=8)
		{
			out[0] = (in[0] + in[4])>>1;
			out[1] = (in[1] + in[5])>>1;
			out[2] = (in[2] + in[6])>>1;
			out[3] = (in[3] + in[7])>>1;
		}
	}
	else if (inwidth <= 1)
	{
		//single colum, peek only at this pixel
		for (i=0 ; i<outheight ; i++, inrow+=rowwidth*2)
		{
			for (in = inrow, j=0 ; j<outwidth ; j++, out+=4, in+=8)
			{
				out[0] = (in[0] + in[rowwidth+0])>>1;
				out[1] = (in[1] + in[rowwidth+1])>>1;
				out[2] = (in[2] + in[rowwidth+2])>>1;
				out[3] = (in[3] + in[rowwidth+3])>>1;
			}
		}
	}
	else
	{
		for (i=0 ; i<outheight ; i++, inrow+=rowwidth*2)
		{
			for (in = inrow, j=0 ; j<outwidth ; j++, out+=4, in+=8)
			{
				out[0] = (in[0] + in[4] + in[rowwidth+0] + in[rowwidth+4])>>2;
				out[1] = (in[1] + in[5] + in[rowwidth+1] + in[rowwidth+5])>>2;
				out[2] = (in[2] + in[6] + in[rowwidth+2] + in[rowwidth+6])>>2;
				out[3] = (in[3] + in[7] + in[rowwidth+3] + in[rowwidth+7])>>2;
			}
		}
	}
}

//oh how I wish I had C++'s template stuff right now
static void Image_MipMap4X16 (unsigned short *in, int inwidth, int inheight, unsigned short *out, int outwidth, int outheight)
{
	int		i, j;
	unsigned short	*inrow;

	int rowwidth = inwidth*4;	//rowwidth is the byte width of the input
	inrow = in;

	//mips round down, except for when the input is 1. which bugs out.
	if (inwidth <= 1 && inheight <= 1)
	{
		out[0] = in[0];
		out[1] = in[1];
		out[2] = in[2];
		out[3] = in[3];
	}
	else if (inheight <= 1)
	{
		//single row, don't peek at the next
		for (in = inrow, j=0 ; j<outwidth ; j++, out+=4, in+=8)
		{
			out[0] = (in[0] + in[4])>>1;
			out[1] = (in[1] + in[5])>>1;
			out[2] = (in[2] + in[6])>>1;
			out[3] = (in[3] + in[7])>>1;
		}
	}
	else if (inwidth <= 1)
	{
		//single colum, peek only at this pixel
		for (i=0 ; i<outheight ; i++, inrow+=rowwidth*2)
		{
			for (in = inrow, j=0 ; j<outwidth ; j++, out+=4, in+=8)
			{
				out[0] = (in[0] + in[rowwidth+0])>>1;
				out[1] = (in[1] + in[rowwidth+1])>>1;
				out[2] = (in[2] + in[rowwidth+2])>>1;
				out[3] = (in[3] + in[rowwidth+3])>>1;
			}
		}
	}
	else
	{
		for (i=0 ; i<outheight ; i++, inrow+=rowwidth*2)
		{
			for (in = inrow, j=0 ; j<outwidth ; j++, out+=4, in+=8)
			{
				out[0] = (in[0] + in[4] + in[rowwidth+0] + in[rowwidth+4])>>2;
				out[1] = (in[1] + in[5] + in[rowwidth+1] + in[rowwidth+5])>>2;
				out[2] = (in[2] + in[6] + in[rowwidth+2] + in[rowwidth+6])>>2;
				out[3] = (in[3] + in[7] + in[rowwidth+3] + in[rowwidth+7])>>2;
			}
		}
	}
}

static void Image_MipMap4X32F (float *in, int inwidth, int inheight, float *out, int outwidth, int outheight)
{
	int		i, j;
	float	*inrow;

	int rowwidth = inwidth*4;	//rowwidth is the byte width of the input
	inrow = in;

	//mips round down, except for when the input is 1. which bugs out.
	if (inwidth <= 1 && inheight <= 1)
	{
		out[0] = in[0];
		out[1] = in[1];
		out[2] = in[2];
		out[3] = in[3];
	}
	else if (inheight <= 1)
	{
		//single row, don't peek at the next
		for (in = inrow, j=0 ; j<outwidth ; j++, out+=4, in+=8)
		{
			out[0] = (in[0] + in[4])/2;
			out[1] = (in[1] + in[5])/2;
			out[2] = (in[2] + in[6])/2;
			out[3] = (in[3] + in[7])/2;
		}
	}
	else if (inwidth <= 1)
	{
		//single colum, peek only at this pixel
		for (i=0 ; i<outheight ; i++, inrow+=rowwidth*2)
		{
			for (in = inrow, j=0 ; j<outwidth ; j++, out+=4, in+=8)
			{
				out[0] = (in[0] + in[rowwidth+0])/2;
				out[1] = (in[1] + in[rowwidth+1])/2;
				out[2] = (in[2] + in[rowwidth+2])/2;
				out[3] = (in[3] + in[rowwidth+3])/2;
			}
		}
	}
	else
	{
		for (i=0 ; i<outheight ; i++, inrow+=rowwidth*2)
		{
			for (in = inrow, j=0 ; j<outwidth ; j++, out+=4, in+=8)
			{
				out[0] = (in[0] + in[4] + in[rowwidth+0] + in[rowwidth+4])/4;
				out[1] = (in[1] + in[5] + in[rowwidth+1] + in[rowwidth+5])/4;
				out[2] = (in[2] + in[6] + in[rowwidth+2] + in[rowwidth+6])/4;
				out[3] = (in[3] + in[7] + in[rowwidth+3] + in[rowwidth+7])/4;
			}
		}
	}
}

static void Image_MipMap1X8 (qbyte *in, int inwidth, int inheight, qbyte *out, int outwidth, int outheight)
{
	int		i, j;
	qbyte	*inrow;

	int rowwidth = inwidth;	//rowwidth is the byte width of the input
	inrow = in;

	//mips round down, except for when the input is 1. which bugs out.
	if (inwidth <= 1 && inheight <= 1)
		out[0] = in[0];
	else if (inheight <= 1)
	{
		//single row, don't peek at the next
		for (in = inrow, j=0 ; j<outwidth ; j++, out+=1, in+=2)
			out[0] = (in[0] + in[1])>>1;
	}
	else if (inwidth <= 1)
	{
		//single colum, peek only at this pixel
		for (i=0 ; i<outheight ; i++, inrow+=rowwidth*2)
			for (in = inrow, j=0 ; j<outwidth ; j++, out+=1, in+=2)
				out[0] = (in[0] + in[rowwidth+0])>>1;
	}
	else
	{
		for (i=0 ; i<outheight ; i++, inrow+=rowwidth*2)
			for (in = inrow, j=0 ; j<outwidth ; j++, out+=1, in+=2)
				out[0] = (in[0] + in[1] + in[rowwidth+0] + in[rowwidth+1])>>2;
	}
}

static qbyte Image_BlendPalette_2(qbyte a, qbyte b)
{
	return a;
}
static qbyte Image_BlendPalette_4(qbyte a, qbyte b, qbyte c, qbyte d)
{
	return a;
}
//this is expected to be slow, thanks to those two expensive helpers.
static void Image_MipMap8Pal (qbyte *in, int inwidth, int inheight, qbyte *out, int outwidth, int outheight)
{
	int		i, j;
	qbyte	*inrow;

	int rowwidth = inwidth;	//rowwidth is the byte width of the input
	inrow = in;

	//mips round down, except for when the input is 1. which bugs out.
	if (inwidth <= 1 && inheight <= 1)
		out[0] = in[0];
	else if (inheight <= 1)
	{
		//single row, don't peek at the next
		for (in = inrow, j=0 ; j<outwidth ; j++, out+=1, in+=2)
			out[0] = Image_BlendPalette_2(in[0], in[1]);
	}
	else if (inwidth <= 1)
	{
		//single colum, peek only at this pixel
		for (i=0 ; i<outheight ; i++, inrow+=rowwidth*2)
			for (in = inrow, j=0 ; j<outwidth ; j++, out+=1, in+=2)
				out[0] = Image_BlendPalette_2(in[0], in[rowwidth]);
	}
	else
	{
		for (i=0 ; i<outheight ; i++, inrow+=rowwidth*2)
			for (in = inrow, j=0 ; j<outwidth ; j++, out+=1, in+=2)
				out[0] = Image_BlendPalette_4(in[0], in[1], in[rowwidth+0], in[rowwidth+1]);
	}
}

static void Image_GenerateMips(struct pendingtextureinfo *mips, unsigned int flags)
{
	int mip;

	if (mips->type != PTI_2D)
		return;	//blurgh

	if (flags & IF_NOMIPMAP)
		return;

	switch(mips->encoding)
	{
	case PTI_P8:
		if (sh_config.can_mipcap)
			return;	//if we can cap mips, do that. it'll save lots of expensive lookups and uglyness.
		for (mip = mips->mipcount; mip < 32; mip++)
		{
			mips->mip[mip].width = mips->mip[mip-1].width >> 1;
			mips->mip[mip].height = mips->mip[mip-1].height >> 1;
			mips->mip[mip].depth = 1;
			if (mips->mip[mip].width < 1 && mips->mip[mip].height < 1)
				break;
			if (mips->mip[mip].width < 1)
				mips->mip[mip].width = 1;
			if (mips->mip[mip].height < 1)
				mips->mip[mip].height = 1;
			mips->mip[mip].datasize = ((mips->mip[mip].width+3)&~3) * mips->mip[mip].height*4;
			mips->mip[mip].data = BZ_Malloc(mips->mip[mip].datasize);
			mips->mip[mip].needfree = true;

			Image_MipMap8Pal(mips->mip[mip-1].data, mips->mip[mip-1].width, mips->mip[mip-1].height, mips->mip[mip].data, mips->mip[mip].width, mips->mip[mip].height);
			mips->mipcount = mip+1;
		}
		return;
	case PTI_R8:
		if (sh_config.can_mipcap)
			return;	//if we can cap mips, do that. it'll save lots of expensive lookups and uglyness.
		for (mip = mips->mipcount; mip < 32; mip++)
		{
			mips->mip[mip].width = mips->mip[mip-1].width >> 1;
			mips->mip[mip].height = mips->mip[mip-1].height >> 1;
			mips->mip[mip].depth = 1;
			if (mips->mip[mip].width < 1 && mips->mip[mip].height < 1)
				break;
			if (mips->mip[mip].width < 1)
				mips->mip[mip].width = 1;
			if (mips->mip[mip].height < 1)
				mips->mip[mip].height = 1;
			mips->mip[mip].datasize = ((mips->mip[mip].width+3)&~3) * mips->mip[mip].height*4;
			mips->mip[mip].data = BZ_Malloc(mips->mip[mip].datasize);
			mips->mip[mip].needfree = true;

			Image_MipMap1X8(mips->mip[mip-1].data, mips->mip[mip-1].width, mips->mip[mip-1].height, mips->mip[mip].data, mips->mip[mip].width, mips->mip[mip].height);
			mips->mipcount = mip+1;
		}
		return;
	case PTI_RGBA32F:
		for (mip = mips->mipcount; mip < 32; mip++)
		{
			mips->mip[mip].width = mips->mip[mip-1].width >> 1;
			mips->mip[mip].height = mips->mip[mip-1].height >> 1;
			mips->mip[mip].depth = 1;
			if (mips->mip[mip].width < 1 && mips->mip[mip].height < 1)
				break;
			if (mips->mip[mip].width < 1)
				mips->mip[mip].width = 1;
			if (mips->mip[mip].height < 1)
				mips->mip[mip].height = 1;
			mips->mip[mip].datasize = ((mips->mip[mip].width+3)&~3) * mips->mip[mip].height * sizeof(float)*4;
			mips->mip[mip].data = BZ_Malloc(mips->mip[mip].datasize);
			mips->mip[mip].needfree = true;

			Image_MipMap4X32F(mips->mip[mip-1].data, mips->mip[mip-1].width, mips->mip[mip-1].height, mips->mip[mip].data, mips->mip[mip].width, mips->mip[mip].height);
			mips->mipcount = mip+1;
		}
		break;
	case PTI_RGBA16:
		for (mip = mips->mipcount; mip < 32; mip++)
		{
			mips->mip[mip].width = mips->mip[mip-1].width >> 1;
			mips->mip[mip].height = mips->mip[mip-1].height >> 1;
			mips->mip[mip].depth = 1;
			if (mips->mip[mip].width < 1 && mips->mip[mip].height < 1)
				break;
			if (mips->mip[mip].width < 1)
				mips->mip[mip].width = 1;
			if (mips->mip[mip].height < 1)
				mips->mip[mip].height = 1;
			mips->mip[mip].datasize = ((mips->mip[mip].width+3)&~3) * mips->mip[mip].height * sizeof(unsigned short)*4;
			mips->mip[mip].data = BZ_Malloc(mips->mip[mip].datasize);
			mips->mip[mip].needfree = true;

			Image_MipMap4X16(mips->mip[mip-1].data, mips->mip[mip-1].width, mips->mip[mip-1].height, mips->mip[mip].data, mips->mip[mip].width, mips->mip[mip].height);
			mips->mipcount = mip+1;
		}
		break;
	case PTI_RGBA8_SRGB:
	case PTI_RGBX8_SRGB:
	case PTI_BGRA8_SRGB:
	case PTI_BGRX8_SRGB:
	case PTI_RGBA8:
	case PTI_RGBX8:
	case PTI_BGRA8:
	case PTI_BGRX8:
		for (mip = mips->mipcount; mip < 32; mip++)
		{
			mips->mip[mip].width = mips->mip[mip-1].width >> 1;
			mips->mip[mip].height = mips->mip[mip-1].height >> 1;
			mips->mip[mip].depth = 1;
			if (mips->mip[mip].width < 1 && mips->mip[mip].height < 1)
				break;
			if (mips->mip[mip].width < 1)
				mips->mip[mip].width = 1;
			if (mips->mip[mip].height < 1)
				mips->mip[mip].height = 1;
			mips->mip[mip].datasize = ((mips->mip[mip].width+3)&~3) * mips->mip[mip].height * sizeof(qbyte)*4;
			mips->mip[mip].data = BZ_Malloc(mips->mip[mip].datasize);
			mips->mip[mip].needfree = true;

			Image_MipMap4X8(mips->mip[mip-1].data, mips->mip[mip-1].width, mips->mip[mip-1].height, mips->mip[mip].data, mips->mip[mip].width, mips->mip[mip].height);
			mips->mipcount = mip+1;
		}
		break;
	case PTI_RGBA4444:
	case PTI_RGB565:
	case PTI_RGBA5551:
		return;	//convert to 16bit afterwards. always mipmap at 8 bit, to try to preserve what little precision there is.
	default:
		return;	//not supported.
	}
}

//stolen from DP
//FIXME: optionally support borders as 0,0,0,0
static void Image_Resample32LerpLine (const qbyte *in, qbyte *out, int inwidth, int outwidth)
{
	int		j, xi, oldx = 0, f, fstep, endx, lerp;
	fstep = (int) (inwidth*65536.0f/outwidth);
	endx = (inwidth-1);
	for (j = 0,f = 0;j < outwidth;j++, f += fstep)
	{
		xi = f >> 16;
		if (xi != oldx)
		{
			in += (xi - oldx) * 4;
			oldx = xi;
		}
		if (xi < endx)
		{
			lerp = f & 0xFFFF;
			*out++ = (qbyte) ((((in[4] - in[0]) * lerp) >> 16) + in[0]);
			*out++ = (qbyte) ((((in[5] - in[1]) * lerp) >> 16) + in[1]);
			*out++ = (qbyte) ((((in[6] - in[2]) * lerp) >> 16) + in[2]);
			*out++ = (qbyte) ((((in[7] - in[3]) * lerp) >> 16) + in[3]);
		}
		else // last pixel of the line has no pixel to lerp to
		{
			*out++ = in[0];
			*out++ = in[1];
			*out++ = in[2];
			*out++ = in[3];
		}
	}
}

//yes, this is lordhavok's code too.
//superblur away!
//FIXME: optionally support borders as 0,0,0,0
#define LERPBYTE(i) r = row1[i];out[i] = (qbyte) ((((row2[i] - r) * lerp) >> 16) + r)
static void Image_Resample32Lerp(const void *indata, int inwidth, int inheight, void *outdata, int outwidth, int outheight)
{
	int i, j, r, yi, oldy, f, fstep, lerp, endy = (inheight-1), inwidth4 = inwidth*4, outwidth4 = outwidth*4;
	qbyte *out;
	const qbyte *inrow;
	qbyte *row1, *row2;

	row1 = alloca(2*(outwidth*4));
	row2 = row1 + (outwidth * 4);

	out = outdata;
	fstep = (int) (inheight*65536.0f/outheight);

	inrow = indata;
	oldy = 0;
	Image_Resample32LerpLine (inrow, row1, inwidth, outwidth);
	Image_Resample32LerpLine (inrow + inwidth4, row2, inwidth, outwidth);
	for (i = 0, f = 0;i < outheight;i++,f += fstep)
	{
		yi = f >> 16;
		if (yi < endy)
		{
			lerp = f & 0xFFFF;
			if (yi != oldy)
			{
				inrow = (const qbyte *)indata + inwidth4*yi;
				if (yi == oldy+1)
					memcpy(row1, row2, outwidth4);
				else
					Image_Resample32LerpLine (inrow, row1, inwidth, outwidth);
				Image_Resample32LerpLine (inrow + inwidth4, row2, inwidth, outwidth);
				oldy = yi;
			}
			j = outwidth - 4;
			while(j >= 0)
			{
				LERPBYTE( 0);
				LERPBYTE( 1);
				LERPBYTE( 2);
				LERPBYTE( 3);
				LERPBYTE( 4);
				LERPBYTE( 5);
				LERPBYTE( 6);
				LERPBYTE( 7);
				LERPBYTE( 8);
				LERPBYTE( 9);
				LERPBYTE(10);
				LERPBYTE(11);
				LERPBYTE(12);
				LERPBYTE(13);
				LERPBYTE(14);
				LERPBYTE(15);
				out += 16;
				row1 += 16;
				row2 += 16;
				j -= 4;
			}
			if (j & 2)
			{
				LERPBYTE( 0);
				LERPBYTE( 1);
				LERPBYTE( 2);
				LERPBYTE( 3);
				LERPBYTE( 4);
				LERPBYTE( 5);
				LERPBYTE( 6);
				LERPBYTE( 7);
				out += 8;
				row1 += 8;
				row2 += 8;
			}
			if (j & 1)
			{
				LERPBYTE( 0);
				LERPBYTE( 1);
				LERPBYTE( 2);
				LERPBYTE( 3);
				out += 4;
				row1 += 4;
				row2 += 4;
			}
			row1 -= outwidth4;
			row2 -= outwidth4;
		}
		else
		{
			yi = endy;	//don't read off the end
			if (yi != oldy)
			{
				inrow = (const qbyte *)indata + inwidth4*yi;
				if (yi == oldy+1)
					memcpy(row1, row2, outwidth4);
				else
					Image_Resample32LerpLine (inrow, row1, inwidth, outwidth);
				oldy = yi;
			}
			memcpy(out, row1, outwidth4);
			out += outwidth4;	//Fixes a bug from DP.
		}
	}
}


/*
================
GL_ResampleTexture
================
*/
static void Image_ResampleTexture32Nearest (const unsigned *in, int inwidth, int inheight, unsigned *out,  int outwidth, int outheight)
{
	int		i, j;
	const unsigned	*inrow;
	unsigned	frac, fracstep;

	fracstep = inwidth*0x10000/outwidth;
	for (i=0 ; i<outheight ; i++, out += outwidth)
	{
		inrow = in + inwidth*(i*inheight/outheight);
		frac = outwidth*fracstep;
		j=outwidth;
		while ((j)&3)
		{
			j--;
			frac -= fracstep;
			out[j] = inrow[frac>>16];
		}
		for ( ; j>=4 ;)
		{
			j-=4;
			frac -= fracstep;
			out[j+3] = inrow[frac>>16];
			frac -= fracstep;
			out[j+2] = inrow[frac>>16];
			frac -= fracstep;
			out[j+1] = inrow[frac>>16];
			frac -= fracstep;
			out[j+0] = inrow[frac>>16];
		}
	}
}

static void Image_ResampleTexture8Nearest (const unsigned char *in, int inwidth, int inheight, unsigned char *out,  int outwidth, int outheight)
{
	int		i, j;
	const unsigned	char *inrow;
	unsigned	frac, fracstep;

	fracstep = inwidth*0x10000/outwidth;
	for (i=0 ; i<outheight ; i++, out += outwidth)
	{
		inrow = in + inwidth*(i*inheight/outheight);
		frac = outwidth*fracstep;
		j=outwidth;
		while ((j)&3)
		{
			j--;
			frac -= fracstep;
			out[j] = inrow[frac>>16];
		}
		for ( ; j>=4 ;)
		{
			j-=4;
			frac -= fracstep;
			out[j+3] = inrow[frac>>16];
			frac -= fracstep;
			out[j+2] = inrow[frac>>16];
			frac -= fracstep;
			out[j+1] = inrow[frac>>16];
			frac -= fracstep;
			out[j+0] = inrow[frac>>16];
		}
	}
}
static void Image_ResampleTexture16Nearest (const unsigned short *in, int inwidth, int inheight, unsigned short *out,  int outwidth, int outheight)
{
	int		i, j;
	const unsigned	short *inrow;
	unsigned	frac, fracstep;

	fracstep = inwidth*0x10000/outwidth;
	for (i=0 ; i<outheight ; i++, out += outwidth)
	{
		inrow = in + inwidth*(i*inheight/outheight);
		frac = outwidth*fracstep;
		j=outwidth;
		while ((j)&3)
		{
			j--;
			frac -= fracstep;
			out[j] = inrow[frac>>16];
		}
		for ( ; j>=4 ;)
		{
			j-=4;
			frac -= fracstep;
			out[j+3] = inrow[frac>>16];
			frac -= fracstep;
			out[j+2] = inrow[frac>>16];
			frac -= fracstep;
			out[j+1] = inrow[frac>>16];
			frac -= fracstep;
			out[j+0] = inrow[frac>>16];
		}
	}
}

//returns out on success. if out is null then returns a new BZ_Malloced buffer
void *Image_ResampleTexture (uploadfmt_t format, const void *in, int inwidth, int inheight, void *out,  int outwidth, int outheight)
{
	switch(format)
	{
	case PTI_INVALID:
	default:
		return NULL;
	//we don't care about byte order etc, just channels+sizes.
	case PTI_LLLX8:
	case PTI_LLLA8:
	case PTI_RGBA8:
	case PTI_RGBX8:
	case PTI_BGRA8:
	case PTI_BGRX8:
	case PTI_RGBA8_SRGB:
	case PTI_RGBX8_SRGB:
	case PTI_BGRA8_SRGB:
	case PTI_BGRX8_SRGB:
		if (!out)
			out = BZ_Malloc(((outwidth+3)&~3)*outheight*4);
		if (gl_lerpimages.ival)
			Image_Resample32Lerp(in, inwidth, inheight, out, outwidth, outheight);	//FIXME: should be sRGB-aware, but probably not a common path on hardware that can actually do srgb.
		else
			Image_ResampleTexture32Nearest(in, inwidth, inheight, out, outwidth, outheight);
		return out;
	case PTI_R32F:
	case PTI_A2BGR10:
	case PTI_E5BGR9:
		if (!out)
			out = BZ_Malloc(((outwidth+3)&~3)*outheight*4);
		Image_ResampleTexture32Nearest(in, inwidth, inheight, out, outwidth, outheight);
		return out;
	case PTI_P8:
	case PTI_L8:
	case PTI_R8:
	case PTI_L8_SRGB:
	case PTI_R8_SNORM:
		if (!out)
			out = BZ_Malloc((outwidth+3)*outheight);
		Image_ResampleTexture8Nearest(in, inwidth, inheight, out, outwidth, outheight);
		return out;
	case PTI_RG8:
	case PTI_RG8_SNORM:
	case PTI_L8A8:
	case PTI_L8A8_SRGB:
	case PTI_R16:
	case PTI_R16F:
	case PTI_RGB565:
	case PTI_RGBA4444:
	case PTI_ARGB4444:
	case PTI_RGBA5551:
	case PTI_ARGB1555:
		if (!out)
			out = BZ_Malloc((outwidth+3)*outheight*2);
		Image_ResampleTexture16Nearest(in, inwidth, inheight, out, outwidth, outheight);
		return out;
	case PTI_RGB8:
	case PTI_BGR8:
	case PTI_RGBA16:
	case PTI_RGBA16F:
	case PTI_RGBA32F:
		return NULL;	//fixmes
	}
}

//ripped from tenebrae
static unsigned int * Image_GenerateNormalMap(qbyte *pixels, unsigned int *nmap, int w, int h, float scale, float offsetscale)
{
	int i, j, wr, hr;
	unsigned char r, g, b, height;
	float sqlen, reciplen, nx, ny, nz;

	const float oneOver255 = 1.0f/255.0f;

	float c, cx, cy, dcx, dcy;

	wr = w;
	hr = h;

	for (i=0; i<h; i++)
	{
		for (j=0; j<w; j++)
		{
			/* Expand [0,255] texel values to the [0,1] range. */
			c = pixels[i*wr + j] * oneOver255;
			/* Expand the texel to its right. */
			cx = pixels[i*wr + (j+1)%wr] * oneOver255;
			/* Expand the texel one up. */
			cy = pixels[((i+1)%hr)*wr + j] * oneOver255;
			dcx = scale * (c - cx);
			dcy = scale * (c - cy);

			/* Normalize the vector. */
			sqlen = dcx*dcx + dcy*dcy + 1;
			reciplen = 1.0f/(float)sqrt(sqlen);
			nx = dcx*reciplen;
			ny = -dcy*reciplen;
			nz = reciplen;

			/* Repack the normalized vector into an RGB unsigned qbyte
			   vector in the normal map image. */
			r = (qbyte) (128 + 127*nx);
			g = (qbyte) (128 + 127*ny);
			b = (qbyte) (128 + 127*nz);

			/* The highest resolution mipmap level always has a
			   unit length magnitude. */
			height = bound(0, (pixels[i*wr + j]*offsetscale)+(255*(1-offsetscale)), 255);
			nmap[i*w+j] = LittleLong((height << 24)|(b << 16)|(g << 8)|(r));	// <AWE> Added support for big endian.
		}
	}

	return &nmap[0];
}

static int Image_GetPicMip(unsigned int flags)
{
	int picmip = 0;
	if (flags & IF_NOMIPMAP)
		picmip += gl_picmip2d.ival;	//2d stuff gets its own picmip cvar.
	else
	{
		picmip += gl_picmip.ival;

		if (flags & IF_WORLDTEX)
			picmip += gl_picmip_world.ival;
		else if (flags & IF_SPRITETEX)
			picmip += gl_picmip_sprites.ival;
		else
			picmip += gl_picmip_other.ival;
	}
	if (picmip < 0)
		picmip = 0;
	return picmip;
}

static void Image_RoundDimensions(int *scaled_width, int *scaled_height, unsigned int flags)
{
	if (sh_config.texture_non_power_of_two)	//NPOT is a simple extension that relaxes errors.
	{
		//lax form
		TRACE(("dbg: GL_RoundDimensions: GL_ARB_texture_non_power_of_two\n"));
	}
	else if ((flags & IF_CLAMP) && (flags & IF_NOMIPMAP) && sh_config.texture_non_power_of_two_pic)
	{
		//more strict form
		TRACE(("dbg: GL_RoundDimensions: GL_OES_texture_npot\n"));
	}
	else
	{
		int width = *scaled_width;
		int height = *scaled_height;
		for (*scaled_width = 1 ; *scaled_width < width ; *scaled_width<<=1)
			;
		for (*scaled_height = 1 ; *scaled_height < height ; *scaled_height<<=1)
			;

		/*round npot textures down if we're running on an embedded system*/
		if (sh_config.npot_rounddown)
		{
			if (*scaled_width != width)
				*scaled_width >>= 1;
			if (*scaled_height != height)
				*scaled_height >>= 1;
		}
	}

	if (!(flags & IF_NOPICMIP))
	{
		int picmip = Image_GetPicMip(flags);
		*scaled_width >>= picmip;
		*scaled_height >>= picmip;
	}

	TRACE(("dbg: GL_RoundDimensions: %f\n", gl_max_size.value));

	if (sh_config.texture2d_maxsize)
	{
		if (*scaled_width > sh_config.texture2d_maxsize)
			*scaled_width = sh_config.texture2d_maxsize;
		if (*scaled_height > sh_config.texture2d_maxsize)
			*scaled_height = sh_config.texture2d_maxsize;
	}
	if (!(flags & (IF_UIPIC|IF_RENDERTARGET)))
	{
		if (gl_max_size.value)
		{
			if (*scaled_width > gl_max_size.value)
				*scaled_width = gl_max_size.value;
			if (*scaled_height > gl_max_size.value)
				*scaled_height = gl_max_size.value;
		}
	}

	if (*scaled_width < 1)
		*scaled_width = 1;
	if (*scaled_height < 1)
		*scaled_height = 1;
}

//may operate in place
static void Image_8888to565(struct pendingtextureinfo *mips, qboolean bgra)
{
	unsigned int mip;
	for (mip = 0; mip < mips->mipcount; mip++)
	{
		qbyte *in = mips->mip[mip].data;
		unsigned short *out = mips->mip[mip].data;
		unsigned int w = mips->mip[mip].width;
		unsigned int h = mips->mip[mip].height;
		unsigned int p = w*h;
		unsigned short tmp;
		if (!mips->mip[mip].needfree && !mips->extrafree)
		{
			mips->mip[mip].needfree = true;
			mips->mip[mip].data = out = BZ_Malloc(sizeof(tmp)*p);
		}

		if (bgra)
		{
			while(p-->0)
			{
				tmp  = ((*in++>>3) << 0);//b
				tmp |= ((*in++>>2) << 5);//g
				tmp |= ((*in++>>3) << 11);//r
				in++;
				*out++ = tmp;
			}
		}
		else
		{
			while(p-->0)
			{
				tmp  = ((*in++>>3) << 11);//r
				tmp |= ((*in++>>2) << 5);//g
				tmp |= ((*in++>>3) << 0);//b
				in++;
				*out++ = tmp;
			}
		}
	}
}

static void Image_8888to1555(struct pendingtextureinfo *mips, qboolean bgra)
{
	unsigned int mip;
	for (mip = 0; mip < mips->mipcount; mip++)
	{
		qbyte *in = mips->mip[mip].data;
		unsigned short *out = mips->mip[mip].data;
		unsigned int w = mips->mip[mip].width;
		unsigned int h = mips->mip[mip].height;
		unsigned int p = w*h;
		unsigned short tmp;
		if (!mips->mip[mip].needfree && !mips->extrafree)
		{
			mips->mip[mip].needfree = true;
			mips->mip[mip].data = out = BZ_Malloc(sizeof(tmp)*p);
		}

		if (bgra)
		{
			while(p-->0)
			{
				tmp  = ((*in++>>3) << 0);//b
				tmp |= ((*in++>>3) << 5);//g
				tmp |= ((*in++>>3) << 10);//r
				tmp |= ((*in++>>7) << 15);//a
				*out++ = tmp;
			}
		}
		else
		{
			while(p-->0)
			{
				tmp  = ((*in++>>3) << 10);//r
				tmp |= ((*in++>>3) << 5);//g
				tmp |= ((*in++>>3) << 0);//b
				tmp |= ((*in++>>7) << 15);//a
				*out++ = tmp;
			}
		}
	}
}

static void Image_8888to5551(struct pendingtextureinfo *mips, qboolean bgra)
{
	unsigned int mip;
	for (mip = 0; mip < mips->mipcount; mip++)
	{
		qbyte *in = mips->mip[mip].data;
		unsigned short *out = mips->mip[mip].data;
		unsigned int w = mips->mip[mip].width;
		unsigned int h = mips->mip[mip].height;
		unsigned int p = w*h;
		unsigned short tmp;
		if (!mips->mip[mip].needfree && !mips->extrafree)
		{
			mips->mip[mip].needfree = true;
			mips->mip[mip].data = out = BZ_Malloc(sizeof(tmp)*p);
		}

		if (bgra)
		{
			while(p-->0)
			{
				tmp  = ((*in++>>3) << 1);//b
				tmp |= ((*in++>>3) << 6);//g
				tmp |= ((*in++>>3) << 11);//r
				tmp |= ((*in++>>7) << 0);//a
				*out++ = tmp;
			}
		}
		else
		{
			while(p-->0)
			{
				tmp  = ((*in++>>3) << 11);//r
				tmp |= ((*in++>>3) << 6);//g
				tmp |= ((*in++>>3) << 1);//b
				tmp |= ((*in++>>7) << 0);//a
				*out++ = tmp;
			}
		}
	}
}

static void Image_8888to4444(struct pendingtextureinfo *mips, qboolean bgra)
{
	unsigned int mip;
	for (mip = 0; mip < mips->mipcount; mip++)
	{
		qbyte *in = mips->mip[mip].data;
		unsigned short *out = mips->mip[mip].data;
		unsigned int w = mips->mip[mip].width;
		unsigned int h = mips->mip[mip].height;
		unsigned int p = w*h;
		unsigned short tmp;
		if (!mips->mip[mip].needfree && !mips->extrafree)
		{
			mips->mip[mip].needfree = true;
			mips->mip[mip].data = out = BZ_Malloc(sizeof(tmp)*p);
		}

		if (bgra)
		{
			while(p-->0)
			{
				tmp  = ((*in++>>4) << 4);//b
				tmp |= ((*in++>>4) << 8);//g
				tmp |= ((*in++>>4) << 12);//r
				tmp |= ((*in++>>4) << 0);//a
				*out++ = tmp;
			}
		}
		else
		{
			while(p-->0)
			{
				tmp  = ((*in++>>4) << 12);//r
				tmp |= ((*in++>>4) << 8);//g
				tmp |= ((*in++>>4) << 4);//b
				tmp |= ((*in++>>4) << 0);//a
				*out++ = tmp;
			}
		}
	}
}
//may operate in place
static void Image_8888toARGB4444(struct pendingtextureinfo *mips, qboolean bgra)
{
	unsigned int mip;
	for (mip = 0; mip < mips->mipcount; mip++)
	{
		qbyte *in = mips->mip[mip].data;
		unsigned short *out = mips->mip[mip].data;
		unsigned int w = mips->mip[mip].width;
		unsigned int h = mips->mip[mip].height;
		unsigned int p = w*h;
		unsigned short tmp;
		if (!mips->mip[mip].needfree && !mips->extrafree)
		{
			mips->mip[mip].needfree = true;
			mips->mip[mip].data = out = BZ_Malloc(sizeof(tmp)*p);
		}

		if (bgra)
		{
			while(p-->0)
			{
				tmp  = ((*in++>>4) << 0);//b
				tmp |= ((*in++>>4) << 4);//g
				tmp |= ((*in++>>4) << 8);//r
				tmp |= ((*in++>>4) << 12);//a
				*out++ = tmp;
			}
		}
		else
		{
			while(p-->0)
			{
				tmp  = ((*in++>>4) << 8);//r
				tmp |= ((*in++>>4) << 4);//g
				tmp |= ((*in++>>4) << 0);//b
				tmp |= ((*in++>>4) << 12);//a
				*out++ = tmp;
			}
		}
	}
}

static void Image_4X16to8888(struct pendingtextureinfo *mips)
{
	unsigned int mip;
	for (mip = 0; mip < mips->mipcount; mip++)
	{
		unsigned short *in = mips->mip[mip].data;
		qbyte *out = mips->mip[mip].data;
		unsigned int w = mips->mip[mip].width;
		unsigned int h = mips->mip[mip].height;
		unsigned int p = w*h*4;
		if (!mips->mip[mip].needfree && !mips->extrafree)
		{
			mips->mip[mip].needfree = true;
			mips->mip[mip].data = out = BZ_Malloc(sizeof(*out)*p);
		}

		while(p-->0)
			*out++ = *in++>>8;
	}
}

//R8,G8,B8,X8 (aligned) -> R8,G8,B8 (tightly packed)
static void Image_32To24(struct pendingtextureinfo *mips)
{
	int mip;
	for (mip = 0; mip < mips->mipcount; mip++)
	{
		qbyte *in = mips->mip[mip].data;
		qbyte *out = mips->mip[mip].data;
		unsigned int w = mips->mip[mip].width;
		unsigned int h = mips->mip[mip].height;
		unsigned int p = w*h;
		if (!mips->mip[mip].needfree && !mips->extrafree)
		{
			mips->mip[mip].needfree = true;
			mips->mip[mip].data = out = BZ_Malloc(sizeof(*out)*p*3);
		}
		while(p-->0)
		{
			*out++ = *in++;
			*out++ = *in++;
			*out++ = *in++;
			in++;
		}
	}
}

//may operate in place
static void Image_8_BGR_RGB_Swap(qbyte *data, unsigned int w, unsigned int h)
{
	unsigned int p = w*h;
	qbyte tmp;
	while(p-->0)
	{
		tmp = data[0];
		data[0] = data[2];
		data[2] = tmp;
		data += 4;
	}
}

typedef union
{
	byte_vec4_t v;
	unsigned int u;
} pixel32_t;
typedef union
{
	unsigned short v[4];
	quint64_t u;
} pixel64_t;
#define etc_expandv(p,x,y,z) p.v[0]|=p.v[0]>>(x),p.v[1]|=p.v[1]>>(y),p.v[2]|=p.v[2]>>(z)
#ifdef DECOMPRESS_ETC2
//FIXME: this is littleendian only...
static void Image_Decode_ETC2_Block_TH_Internal(qbyte *fte_restrict in, pixel32_t *fte_restrict out, int w, pixel32_t base1, pixel32_t base2, int d, qboolean tmode)
{
	pixel32_t painttable[4];
	int dtab[] = {3,6,11,16,23,32,41,64};	//writing that felt like giving out lottery numbers.
#define etc_addclamptopixel(r,p,d) r.v[0]=bound(0,p.v[0]+d, 255),r.v[1]=bound(0,p.v[1]+d, 255),r.v[2]=bound(0,p.v[2]+d, 255),r.v[3]=0xff
	d = dtab[d];
	if (tmode)
	{
		painttable[0].u = base1.u;
		etc_addclamptopixel(painttable[1], base2, d);
		painttable[2].u = base2.u;
		etc_addclamptopixel(painttable[3], base2, -d);
	}
	else
	{
		etc_addclamptopixel(painttable[0], base1, d);
		etc_addclamptopixel(painttable[1], base1, -d);
		etc_addclamptopixel(painttable[2], base2, d);
		etc_addclamptopixel(painttable[3], base2, -d);
	}
#undef etc_addclamptoint
	//yay, we have our painttable. now use it. also, screw that msb/lsb split.

#define etc2_th_pix(r,i)								\
	if (in[5-(i/8)]&(1<<(i&7))) {						\
		if (in[7-(i/8)]&(1<<(i&7)))	r.u = painttable[3].u;	\
		else						r.u = painttable[2].u;	\
	} else {											\
		if (in[7-(i/8)]&(1<<(i&7)))	r.u = painttable[1].u;	\
		else						r.u = painttable[0].u;	\
	}

	etc2_th_pix(out[0], 0);
	etc2_th_pix(out[1], 4);
	etc2_th_pix(out[2], 8);
	etc2_th_pix(out[3], 12);
	out += w;
	etc2_th_pix(out[0], 1);
	etc2_th_pix(out[1], 5);
	etc2_th_pix(out[2], 9);
	etc2_th_pix(out[3], 13);
	out += w;
	etc2_th_pix(out[0], 2);
	etc2_th_pix(out[1], 6);
	etc2_th_pix(out[2], 10);
	etc2_th_pix(out[3], 14);
	out += w;
	etc2_th_pix(out[0], 3);
	etc2_th_pix(out[1], 7);
	etc2_th_pix(out[2], 11);
	etc2_th_pix(out[3], 15);
#undef etc2_th_pix
}
static void Image_Decode_ETC2_Block_Internal(qbyte *fte_restrict in, pixel32_t *fte_restrict out0, int w, int alphamode)
{
	//the overflow modes are only valid with ETC2.
	//alphamode=1 is used for punchthrough-alpha (which also forces the diff mode)
	static const char tab[8][2] =
	{
		{2,8},
		{5,17},
		{9,29},
		{13,42},
		{18,60},
		{24,80},
		{33,106},
		{47,183}
	};
	int tv;
	pixel32_t base1, base2, base3;
	const char *cw1, *cw2;
	unsigned char R1,G1,B1;
	pixel32_t *out1, *out2, *out3;

	qboolean opaque;

	if (alphamode)
		opaque = in[3]&2;
	else
		opaque = 1;

	if (alphamode || (in[3]&2))	//diffbit, bit 33
	{
		R1=(in[0]>>3)&31;//59+5
		G1=(in[1]>>3)&31;//51+5
		B1=(in[2]>>3)&31;//43+5
		VectorSet(base1.v, (R1<<3)+(R1>>2), (G1<<3)+(G1>>2), (B1<<3)+(B1>>2));
		R1 += (char)((in[0]&3)|((in[0]&4)*0x3f));	//56+3
		if (R1&~0x1f) //R2 overflow = T mode
		{
			Vector4Set(base1.v, ((in[0]&0x18)<<3) | ((in[0]&0x3)<<4), (in[1]&0xf0), ((in[1]&0x0f)<<4), 0xff);
			Vector4Set(base2.v, (in[2]&0xf0), ((in[2]&0x0f)<<4), (in[3]&0xf0), 0xff);
			tv = ((in[3]&0x0c)>>1)|(in[3]&0x01);
			etc_expandv(base1,4,4,4);
			etc_expandv(base2,4,4,4);
			Image_Decode_ETC2_Block_TH_Internal(in, out0, w, base1, base2, tv, true);
			return;
		}
		G1 += (char)((in[1]&3)|((in[1]&4)*0x3f));	//48+3
		if (G1&~0x1f) //G2 overflow = H mode
		{
			Vector4Set(base1.v, ((in[0]&0x78)<<1), ((in[0]&0x07)<<5)|((in[1]&0x10)<<0), ((in[1]&0x08)<<4)|((in[1]&0x03)<<5)|((in[2]&0x80)>>3), 0xff);
			Vector4Set(base2.v, ((in[2]&0x78)<<1), ((in[2]&0x07)<<5)|((in[3]&0x80)>>3), ((in[3]&0x78)<<1), 0xff);
			tv = ((in[3]&0x04)>>1)|(in[3]&0x01);
			etc_expandv(base1,4,4,4);
			etc_expandv(base2,4,4,4);
			Image_Decode_ETC2_Block_TH_Internal(in, out0, w, base1, base2, tv, false);
			return;
		}
		B1 += (char)((in[2]&3)|((in[2]&4)*0x3f));	//40+3
		if (B1&~0x1f) //B2 overflow = Planar mode
		{//origin horizontal, vertical delas, interpolated across the 16 pixels
			VectorSet(base1.v, ((in[0]&0x7f)<<1),((in[0]&0x01)<<7)|((in[1])&0x7e),(in[1]<<7)|((in[2]&0x18)<<2)|((in[2]&0x3)<<3)|((in[3]&0x80)>>5));
			VectorSet(base2.v, ((in[3]&0x7c)<<1)|((in[3]&0x01)<<2),(in[4]&0xfe),((in[4]&1)<<7)|((in[5]&0xf8)>>1));
			VectorSet(base3.v, ((in[5]&0x07)<<5)|((in[6]&0xe0)>>3),((in[6]&0x1f)<<3)|((in[7]&0xc0)>>5),(in[7]&0x3f)<<2);
			etc_expandv(base1,6,7,6);
			etc_expandv(base2,6,7,6);
			etc_expandv(base3,6,7,6);
#define etc2_planar2(r,x,y)				\
	r[x].v[0] =	bound(0,(4*base1.v[0] + x*((short)base2.v[0]-base1.v[0]) + y*((short)base3.v[0]-base1.v[0]) + 2)>>2,0xff),	\
	r[x].v[1] = bound(0,(4*base1.v[1] + x*((short)base2.v[1]-base1.v[1]) + y*((short)base3.v[1]-base1.v[1]) + 2)>>2,0xff),	\
	r[x].v[2] = bound(0,(4*base1.v[2] + x*((short)base2.v[2]-base1.v[2]) + y*((short)base3.v[2]-base1.v[2]) + 2)>>2,0xff),	\
	r[x].v[3] = 0xff
#define etc2_planar(r,y)				\
					etc2_planar2(r,0,y);		\
					etc2_planar2(r,1,y);		\
					etc2_planar2(r,2,y);		\
					etc2_planar2(r,3,y);
			etc2_planar(out0,0);out0 += w;
			etc2_planar(out0,1);out0 += w;
			etc2_planar(out0,2);out0 += w;
			etc2_planar(out0,3);
			return;
		}
		//they should still be 5 bits.
		VectorSet(base2.v, (R1<<3)+(R1>>2), (G1<<3)+(G1>>2), (B1<<3)+(B1>>2));
	}
	else
	{
		VectorSet(base1.v, ((in[0]>>4)&15)*0x11,	/*60+4*/
						 ((in[1]>>4)&15)*0x11,	/*52+4*/
						 ((in[2]>>4)&15)*0x11);	/*44+4*/
		VectorSet(base2.v, ((in[0]>>0)&15)*0x11,	/*56+4*/
						 ((in[1]>>0)&15)*0x11,	/*48+4*/
						 ((in[2]>>0)&15)*0x11);	/*40+4*/
	}

	cw1 = tab[(in[3]>>5)&7];	//37+3
	cw2 = tab[(in[3]>>2)&7];	//34+3

	out1 = out0+w*1;
	out2 = out0+w*2;
	out3 = out0+w*3;

#define etc1_pix(r, base,cw,i)	\
	if (in[7-(i/8)]&(1<<(i&7)))						\
		tv = (in[5-(i/8)]&(1<<(i&7)))?-cw[1]:cw[1];	\
	else if (opaque)								\
		tv = (in[5-(i/8)]&(1<<(i&7)))?-cw[0]:cw[0];	\
	else /*punchthrough alpha mode*/				\
		tv = (in[5-(i/8)]&(1<<(i&7)))?-255:0;		\
	if (tv==-255)									\
		r.u = 0;									\
	else											\
		r.v[0] = bound(0,base.v[0]+tv,0xff),			\
		r.v[1] = bound(0,base.v[1]+tv,0xff),			\
		r.v[2] = bound(0,base.v[2]+tv,0xff),			\
		r.v[3] = 0xff

	etc1_pix(out0[0], base1,cw1,0);
	etc1_pix(out0[1], base1,cw1,4);
	etc1_pix(out1[0], base1,cw1,1);
	etc1_pix(out1[1], base1,cw1,5);

	etc1_pix(out2[2], base2,cw2,10);
	etc1_pix(out2[3], base2,cw2,14);
	etc1_pix(out3[2], base2,cw2,11);
	etc1_pix(out3[3], base2,cw2,15);
	if (in[3]&1)	//flipbit bit 32 - blocks are vertical
	{
		etc1_pix(out0[2], base1,cw1,8);
		etc1_pix(out0[3], base1,cw1,12);
		etc1_pix(out1[2], base1,cw1,9);
		etc1_pix(out1[3], base1,cw1,13);

		etc1_pix(out2[0], base2,cw2,2);
		etc1_pix(out2[1], base2,cw2,6);
		etc1_pix(out3[0], base2,cw2,3);
		etc1_pix(out3[1], base2,cw2,7);
	}
	else
	{
		etc1_pix(out0[2], base2,cw2,8);
		etc1_pix(out0[3], base2,cw2,12);
		etc1_pix(out1[2], base2,cw2,9);
		etc1_pix(out1[3], base2,cw2,13);

		etc1_pix(out2[0], base1,cw1,2);
		etc1_pix(out2[1], base1,cw1,6);
		etc1_pix(out3[0], base1,cw1,3);
		etc1_pix(out3[1], base1,cw1,7);
	}
#undef etc1_pix
}
static void Image_Decode_EAC8U_Block_Internal(qbyte *fte_restrict in, qbyte *fte_restrict out, int stride, qboolean goestoeleven)
{
	static const char tabs[16][8] =
	{
		{-3,-6, -9,-15,2,5,8,14},
		{-3,-7,-10,-13,2,6,9,12},
		{-2,-5, -8,-13,1,4,7,12},
		{-2,-4, -6,-13,1,3,5,12},
		{-3,-6, -8,-12,2,5,7,11},
		{-3,-7, -9,-11,2,6,8,10},
		{-4,-7, -8,-11,3,6,7,10},
		{-3,-5, -8,-11,2,4,7,10},
		{-2,-6, -8,-10,1,5,7,9},
		{-3,-5, -8,-10,1,4,7,9},
		{-2,-4, -8,-10,1,3,7,9},
		{-2,-5, -7,-10,1,4,6,9},
		{-3,-4, -7,-10,2,3,6,9},
		{-1,-2, -3,-10,0,1,2,9},
		{-4,-6, -8, -9,3,5,7,8},
		{-3,-5, -7, -9,2,4,6,8},
	};

	const qbyte base = in[0];
	const qbyte mul = in[1]>>4;
	const char *tab = tabs[in[1]&0xf];
	const quint64_t bits = in[2] | (in[3]<<8) | (in[4]<<16) | (in[5]<<24) | ((quint64_t)in[6]<<32) | ((quint64_t)in[7]<<40);

#define EAC_Pix(r,x,y)	r = bound(0, base + tab[(bits>>((x*4+y)*3))&7] * mul, 255);
#define EAC_Row(y)	EAC_Pix(out[0], 0,y);EAC_Pix(out[1], 1,y);EAC_Pix(out[2], 2,y);EAC_Pix(out[3], 3,y);
	EAC_Row(0);out += stride;EAC_Row(1);out += stride;EAC_Row(2);out += stride;EAC_Row(3);
#undef EAC_Row
#undef EAC_Pix
}
static void Image_Decode_ETC2_RGB8_Block(qbyte *fte_restrict in, pixel32_t *fte_restrict out, int w)
{
	Image_Decode_ETC2_Block_Internal(in, out, w, false);
}
//punchthrough alpha works by removing interleaved mode releasing a bit that says whether a block can have alpha=0, .
static void Image_Decode_ETC2_RGB8A1_Block(qbyte *fte_restrict in, pixel32_t *fte_restrict out, int w)
{
	Image_Decode_ETC2_Block_Internal(in, out, w, true);
}
//ETC2 RGBA's alpha and R11(and RG11) work the same way as each other, but with varying extra blocks with either 8 or 11 bits of valid precision.
static void Image_Decode_ETC2_RGB8A8_Block(qbyte *fte_restrict in, pixel32_t *fte_restrict out, int w)
{
	Image_Decode_ETC2_Block_Internal(in+8, out, w, false);
	Image_Decode_EAC8U_Block_Internal(in, out->v+3, w*4, false);
}
static void Image_Decode_EAC_R11U_Block(qbyte *fte_restrict in, pixel32_t *fte_restrict out, int w)
{
	pixel32_t r;
	int i = 0;
	Vector4Set(r.v, 0, 0, 0, 0xff);
	for (i = 0; i < 4; i++)
		out[w*0+i] = out[w*1+i] = out[w*2+i] = out[w*3+i] = r;
	Image_Decode_EAC8U_Block_Internal(in, out->v, w*4, false);
}
static void Image_Decode_EAC_RG11U_Block(qbyte *fte_restrict in, pixel32_t *fte_restrict out, int w)
{
	pixel32_t r;
	int i = 0;
	Vector4Set(r.v, 0, 0, 0, 0xff);
	for (i = 0; i < 4; i++)
		out[w*0+i] = out[w*1+i] = out[w*2+i] = out[w*3+i] = r;
	Image_Decode_EAC8U_Block_Internal(in, out->v+0, w*4, false);
	Image_Decode_EAC8U_Block_Internal(in+8, out->v+1, w*4, false);
}
#endif

#ifdef DECOMPRESS_S3TC
static void Image_Decode_S3TC_Block_Internal(qbyte *fte_restrict in, pixel32_t *fte_restrict out, int w, qbyte blackalpha)
{
	pixel32_t tab[4];
	unsigned int bits;

	Vector4Set(tab[0].v, (in[1]&0xf8), ((in[0]&0xe0)>>3)|((in[1]&7)<<5), (in[0]&0x1f)<<3, 0xff);
	etc_expandv(tab[0],5,6,5);
	Vector4Set(tab[1].v, (in[3]&0xf8), ((in[2]&0xe0)>>3)|((in[3]&7)<<5), (in[2]&0x1f)<<3, 0xff);
	etc_expandv(tab[1],5,6,5);

#define BC1_Lerp(a,as,b,bs,div,c)		((c)[0]=((a)[0]*(as)+(b)[0]*(bs))/(div),(c)[1]=((a)[1]*(as)+(b)[1]*(bs))/(div), (c)[2]=((a)[2]*(as)+(b)[2]*(bs))/(div), (c)[3] = 0xff)
	if ((in[0]|(in[1]<<8)) > (in[2]|(in[3]<<8)))
	{
		BC1_Lerp(tab[0].v,2, tab[1].v,1, 3,tab[2].v);
		BC1_Lerp(tab[0].v,1, tab[1].v,2, 3,tab[3].v);
	}
	else
	{
		BC1_Lerp(tab[0].v,1, tab[1].v,1, 2,tab[2].v);
		Vector4Set(tab[3].v, 0, 0, 0, blackalpha);
	}

	bits = in[4] | (in[5]<<8) | (in[6]<<16) | (in[7]<<24);

#define BC1_Pix(r,i)	r.u = tab[(bits>>((i)*2))&3].u
#define BC1_Row(r,i)	BC1_Pix(r[0],i+0);BC1_Pix(r[1],i+1);BC1_Pix(r[2],i+2);BC1_Pix(r[3],i+3)

	BC1_Row(out, 0);
	out += w;
	BC1_Row(out, 4);
	out += w;
	BC1_Row(out, 8);
	out += w;
	BC1_Row(out, 12);
}
static void Image_Decode_BC1_Block(qbyte *fte_restrict in, pixel32_t *fte_restrict out, int w)
{
	Image_Decode_S3TC_Block_Internal(in, out, w, 0xff);
}
static void Image_Decode_BC1A_Block(qbyte *fte_restrict in, pixel32_t *fte_restrict out, int w)
{
	Image_Decode_S3TC_Block_Internal(in, out, w, 0);
}
static void Image_Decode_BC2_Block(qbyte *fte_restrict in, pixel32_t *fte_restrict out, int w)
{
	Image_Decode_S3TC_Block_Internal(in+8, out, w, 0xff);

	//BC2 has straight 4-bit alpha.
#define BC2_AlphaRow()	\
	out[0].v[3] = in[0]&0x0f; out[0].v[3] |= out[0].v[3]<<4;	\
	out[1].v[3] = in[0]&0xf0; out[1].v[3] |= out[1].v[3]>>4;	\
	out[2].v[3] = in[1]&0x0f; out[2].v[3] |= out[2].v[3]<<4;	\
	out[3].v[3] = in[1]&0xf0; out[3].v[3] |= out[3].v[3]>>4

	BC2_AlphaRow();
	in += 2;out += w;
	BC2_AlphaRow();
	in += 2;out += w;
	BC2_AlphaRow();
	in += 2;out += w;
	BC2_AlphaRow();
#undef BC2_AlphaRow
}
#endif
#ifdef DECOMPRESS_RGTC
static void Image_Decode_RGTC_Block_Internal(qbyte *fte_restrict in, qbyte *fte_restrict out, int stride, qboolean issigned)
{
	quint64_t bits;
	union
	{
		qbyte u;
		char s;
	} tab[8];
	tab[0].u = in[0];
	tab[1].u = in[1];
	if (issigned)
	{
		if (tab[0].s > tab[1].s)
		{
			tab[2].s = (tab[0].s*6 + tab[1].s*1)/7;
			tab[3].s = (tab[0].s*5 + tab[1].s*2)/7;
			tab[4].s = (tab[0].s*4 + tab[1].s*3)/7;
			tab[5].s = (tab[0].s*3 + tab[1].s*4)/7;
			tab[6].s = (tab[0].s*2 + tab[1].s*5)/7;
			tab[7].s = (tab[0].s*1 + tab[1].s*6)/7;
		}
		else
		{
			tab[2].s = (tab[0].s*4 + tab[1].s*1)/5;
			tab[3].s = (tab[0].s*3 + tab[1].s*2)/5;
			tab[4].s = (tab[0].s*2 + tab[1].s*3)/5;
			tab[5].s = (tab[0].s*1 + tab[1].s*4)/5;
			tab[6].s = -128;
			tab[7].s = 127;
		}
	}
	else
	{
		if (tab[0].u > tab[1].u)
		{
			tab[2].u = (tab[0].u*6 + tab[1].u*1)/7;
			tab[3].u = (tab[0].u*5 + tab[1].u*2)/7;
			tab[4].u = (tab[0].u*4 + tab[1].u*3)/7;
			tab[5].u = (tab[0].u*3 + tab[1].u*4)/7;
			tab[6].u = (tab[0].u*2 + tab[1].u*5)/7;
			tab[7].u = (tab[0].u*1 + tab[1].u*6)/7;
		}
		else
		{
			tab[2].u = (tab[0].u*4 + tab[1].u*1)/5;
			tab[3].u = (tab[0].u*3 + tab[1].u*2)/5;
			tab[4].u = (tab[0].u*2 + tab[1].u*3)/5;
			tab[5].u = (tab[0].u*1 + tab[1].u*4)/5;
			tab[6].u = 0;
			tab[7].u = 0xff;
		}
	}

	bits = in[2] | (in[3]<<8) | (in[4]<<16) | (in[5]<<24) | ((quint64_t)in[6]<<32) | ((quint64_t)in[7]<<40);

#define BC3AU_Pix(r,i)	r = tab[(bits>>((i)*3))&7].u;
#define BC3AU_Row(i)	BC3AU_Pix(out[0], i+0);BC3AU_Pix(out[4], i+1);BC3AU_Pix(out[8], i+2);BC3AU_Pix(out[12], i+3);
	BC3AU_Row(0);out += stride;BC3AU_Row(4);out += stride;BC3AU_Row(8);out += stride;BC3AU_Row(12);
#undef BC3AU_Pix
}
#ifdef DECOMPRESS_S3TC
//s3tc rgb channel, with an rgtc alpha channel that depends upon both encodings (really the origin of rgtc, but mneh).
static void Image_Decode_BC3_Block(qbyte *fte_restrict in, pixel32_t *fte_restrict out, int w)
{
	Image_Decode_S3TC_Block_Internal(in+8, out, w, 0xff);
	Image_Decode_RGTC_Block_Internal(in, out->v+3, w*4, false);
}
#endif
static void Image_Decode_BC4U_Block(qbyte *fte_restrict in, pixel32_t *fte_restrict out, int w)
{	//BC4: BC3's alpha channel but used as red only.
	pixel32_t r;
	int i = 0;
	Vector4Set(r.v, 0, 0, 0, 0xff);
	for (i = 0; i < 4; i++)
		out[w*0+i] = out[w*1+i] = out[w*2+i] = out[w*3+i] = r;
	Image_Decode_RGTC_Block_Internal(in, out->v+0, w*4, false);
}
static void Image_Decode_BC4S_Block(qbyte *fte_restrict in, pixel32_t *fte_restrict out, int w)
{	//BC4: BC3's alpha channel but used as red only.
	pixel32_t r;
	int i = 0;
	Vector4Set(r.v, 0, 0, 0, 0xff);
	for (i = 0; i < 4; i++)
		out[w*0+i] = out[w*1+i] = out[w*2+i] = out[w*3+i] = r;
	Image_Decode_RGTC_Block_Internal(in, out->v+0, w*4, true);
}
static void Image_Decode_BC5U_Block(qbyte *fte_restrict in, pixel32_t *fte_restrict out, int w)
{	//BC5: two of BC3's alpha channels but used as red+green only.
	pixel32_t r;
	int i = 0;
	Vector4Set(r.v, 0, 0, 0, 0xff);
	for (i = 0; i < 4; i++)
		out[w*0+i] = out[w*1+i] = out[w*2+i] = out[w*3+i] = r;
	Image_Decode_RGTC_Block_Internal(in+0, out->v+0, w*4, false);
	Image_Decode_RGTC_Block_Internal(in+8, out->v+1, w*4, false);
}
static void Image_Decode_BC5S_Block(qbyte *fte_restrict in, pixel32_t *fte_restrict out, int w)
{	//BC5: two of BC3's alpha channels but used as red+green only.
	pixel32_t r;
	int i = 0;
	Vector4Set(r.v, 0, 0, 0, 0xff);
	for (i = 0; i < 4; i++)
		out[w*0+i] = out[w*1+i] = out[w*2+i] = out[w*3+i] = r;
	Image_Decode_RGTC_Block_Internal(in+0, out->v+0, w*4, true);
	Image_Decode_RGTC_Block_Internal(in+8, out->v+1, w*4, true);
}
#endif

#ifdef DECOMPRESS_BPTC
fte_inlinestatic int ReadBits(qbyte *in, int *bit, int n)
{
	int mask = (1<<n)-1;
	int second;
	int first = *bit;
	*bit += n;
	in += first>>3;
	first &= 7;
	second = max(0,n - (8-first));
	return ((in[0]>>first)&mask) ^ (in[1]<<(n-second)&mask);
}
fte_inlinestatic int ReadBitsL(qbyte *in, int *bit, int n)
{
	int r = ReadBits(in, bit, 8);
	return (r)|(ReadBits(in, bit, n-8)<<8);
}
static void Image_Decode_BC6_Block(qbyte *fte_restrict in, pixel64_t *fte_restrict out, int w, qboolean signextend)
{
	static const int anchors[32] =
	{
		15,15,15,15,
		15,15,15,15,
		15,15,15,15,
		15,15,15,15,
		15, 2, 8, 2,
		2, 8, 8,15,
		2, 8, 2, 2,
		8, 8, 2, 2,
	};
	static const qbyte p2[] = {
		0,0,1,1,0,0,1,1,0,0,1,1,0,0,1,1,
		0,0,0,1,0,0,0,1,0,0,0,1,0,0,0,1,
		0,1,1,1,0,1,1,1,0,1,1,1,0,1,1,1,
		0,0,0,1,0,0,1,1,0,0,1,1,0,1,1,1,
		0,0,0,0,0,0,0,1,0,0,0,1,0,0,1,1,
		0,0,1,1,0,1,1,1,0,1,1,1,1,1,1,1,
		0,0,0,1,0,0,1,1,0,1,1,1,1,1,1,1,
		0,0,0,0,0,0,0,1,0,0,1,1,0,1,1,1,
		0,0,0,0,0,0,0,0,0,0,0,1,0,0,1,1,
		0,0,1,1,0,1,1,1,1,1,1,1,1,1,1,1,
		0,0,0,0,0,0,0,1,0,1,1,1,1,1,1,1,
		0,0,0,0,0,0,0,0,0,0,0,1,0,1,1,1,
		0,0,0,1,0,1,1,1,1,1,1,1,1,1,1,1,
		0,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,
		0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,1,
		0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,
		0,0,0,0,1,0,0,0,1,1,1,0,1,1,1,1,
		0,1,1,1,0,0,0,1,0,0,0,0,0,0,0,0,
		0,0,0,0,0,0,0,0,1,0,0,0,1,1,1,0,
		0,1,1,1,0,0,1,1,0,0,0,1,0,0,0,0,
		0,0,1,1,0,0,0,1,0,0,0,0,0,0,0,0,
		0,0,0,0,1,0,0,0,1,1,0,0,1,1,1,0,
		0,0,0,0,0,0,0,0,1,0,0,0,1,1,0,0,
		0,1,1,1,0,0,1,1,0,0,1,1,0,0,0,1,
		0,0,1,1,0,0,0,1,0,0,0,1,0,0,0,0,
		0,0,0,0,1,0,0,0,1,0,0,0,1,1,0,0,
		0,1,1,0,0,1,1,0,0,1,1,0,0,1,1,0,
		0,0,1,1,0,1,1,0,0,1,1,0,1,1,0,0,
		0,0,0,1,0,1,1,1,1,1,1,0,1,0,0,0,
		0,0,0,0,1,1,1,1,1,1,1,1,0,0,0,0,
		0,1,1,1,0,0,0,1,1,0,0,0,1,1,1,0,
		0,0,1,1,1,0,0,1,1,0,0,1,1,1,0,0,
		0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1,
		0,0,0,0,1,1,1,1,0,0,0,0,1,1,1,1,
		0,1,0,1,1,0,1,0,0,1,0,1,1,0,1,0,
		0,0,1,1,0,0,1,1,1,1,0,0,1,1,0,0,
		0,0,1,1,1,1,0,0,0,0,1,1,1,1,0,0,
		0,1,0,1,0,1,0,1,1,0,1,0,1,0,1,0,
		0,1,1,0,1,0,0,1,0,1,1,0,1,0,0,1,
		0,1,0,1,1,0,1,0,1,0,1,0,0,1,0,1,
		0,1,1,1,0,0,1,1,1,1,0,0,1,1,1,0,
		0,0,0,1,0,0,1,1,1,1,0,0,1,0,0,0,
		0,0,1,1,0,0,1,0,0,1,0,0,1,1,0,0,
		0,0,1,1,1,0,1,1,1,1,0,1,1,1,0,0,
		0,1,1,0,1,0,0,1,1,0,0,1,0,1,1,0,
		0,0,1,1,1,1,0,0,1,1,0,0,0,0,1,1,
		0,1,1,0,0,1,1,0,1,0,0,1,1,0,0,1,
		0,0,0,0,0,1,1,0,0,1,1,0,0,0,0,0,
		0,1,0,0,1,1,1,0,0,1,0,0,0,0,0,0,
		0,0,1,0,0,1,1,1,0,0,1,0,0,0,0,0,
		0,0,0,0,0,0,1,0,0,1,1,1,0,0,1,0,
		0,0,0,0,0,1,0,0,1,1,1,0,0,1,0,0,
		0,1,1,0,1,1,0,0,1,0,0,1,0,0,1,1,
		0,0,1,1,0,1,1,0,1,1,0,0,1,0,0,1,
		0,1,1,0,0,0,1,1,1,0,0,1,1,1,0,0,
		0,0,1,1,1,0,0,1,1,1,0,0,0,1,1,0,
		0,1,1,0,1,1,0,0,1,1,0,0,1,0,0,1,
		0,1,1,0,0,0,1,1,0,0,1,1,1,0,0,1,
		0,1,1,1,1,1,1,0,1,0,0,0,0,0,0,1,
		0,0,0,1,1,0,0,0,1,1,1,0,0,1,1,1,
		0,0,0,0,1,1,1,1,0,0,1,1,0,0,1,1,
		0,0,1,1,0,0,1,1,1,1,1,1,0,0,0,0,
		0,0,1,0,0,0,1,0,1,1,1,0,1,1,1,0,
		0,1,0,0,0,1,0,0,0,1,1,1,0,1,1,1,
    };
    static const int w3[] = {0, 9, 18, 27, 37, 46, 55, 64};
    static const int w4[] = {0, 4, 9, 13, 17, 21, 26, 30, 34, 38, 43, 47, 51, 55, 60, 64};
    static const int *wsz[] = {w3,w3,w3, w3,w4};
    const int *weight;

	unsigned short rgb[4][3] = {{0}};
	unsigned short palette[16][3];
	int i, j, cb, ib;
	int bit = 0, cbits[4];
	int ss;
	int shapeindex;
	int tr;
	int mode = ReadBits(in, &bit, 2);
	if (mode >= 2)
		mode |=ReadBits(in, &bit, 3)<<2;

	switch(mode)
	{
	case 0:	//1
		Vector4Set(cbits, 5,5,5, 10);
		tr = 1;
		ss = 2;
		rgb[2][1] |= ReadBits(in, &bit, 1)<<4;	//g2[4],
		rgb[2][2] |= ReadBits(in, &bit, 1)<<4;	//b2[4],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<4;	//b3[4],
		rgb[0][0] |= ReadBitsL(in, &bit,10)<<0;	//r0[9:0],
		rgb[0][1] |= ReadBitsL(in, &bit,10)<<0;	//g0[9:0],
		rgb[0][2] |= ReadBitsL(in, &bit,10)<<0;	//b0[9:0],
		rgb[1][0] |= ReadBits(in, &bit, 5)<<0;	//r1[4:0],
		rgb[3][1] |= ReadBits(in, &bit, 1)<<4;	//g3[4],
		rgb[2][1] |= ReadBits(in, &bit, 4)<<0;	//g2[3:0],
		rgb[1][1] |= ReadBits(in, &bit, 5)<<0;	//g1[4:0],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<0;	//b3[0],
		rgb[3][1] |= ReadBits(in, &bit, 4)<<0;	//g3[3:0],
		rgb[1][2] |= ReadBits(in, &bit, 5)<<0;	//b1[4:0],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<1;	//b3[1],
		rgb[2][2] |= ReadBits(in, &bit, 4)<<0;	//b2[3:0],
		rgb[2][0] |= ReadBits(in, &bit, 5)<<0;	//r2[4:0],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<2;	//b3[2],
		rgb[3][0] |= ReadBits(in, &bit, 5)<<0;	//r3[4:0],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<3;	//b3[3]
		shapeindex = ReadBits(in, &bit, 5);
		break;
	case 1:	//2
		Vector4Set(cbits, 6,6,6, 7);
		tr = 1;
		ss = 2;
		rgb[2][1] |= ReadBits(in, &bit, 1)<<5;	//g2[5],
		rgb[3][1] |= ReadBits(in, &bit, 1)<<4;	//g3[4],
		rgb[3][1] |= ReadBits(in, &bit, 1)<<5;	//g3[5],
		rgb[0][0] |= ReadBits(in, &bit, 7)<<0;	//r0[6:0],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<0;	//b3[0],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<1;	//b3[1],
		rgb[2][2] |= ReadBits(in, &bit, 1)<<4;	//b2[4],
		rgb[0][1] |= ReadBits(in, &bit, 7)<<0;	//g0[6:0],
		rgb[2][2] |= ReadBits(in, &bit, 1)<<5;	//b2[5],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<2;	//b3[2],
		rgb[2][1] |= ReadBits(in, &bit, 1)<<4;	//g2[4],
		rgb[0][2] |= ReadBits(in, &bit, 7)<<0;	//b0[6:0],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<3;	//b3[3],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<5;	//b3[5],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<4;	//b3[4],
		rgb[1][0] |= ReadBits(in, &bit, 6)<<0;	//r1[5:0],
		rgb[2][1] |= ReadBits(in, &bit, 4)<<0;	//g2[3:0],
		rgb[1][1] |= ReadBits(in, &bit, 6)<<0;	//g1[5:0],
		rgb[3][1] |= ReadBits(in, &bit, 4)<<0;	//g3[3:0],
		rgb[1][2] |= ReadBits(in, &bit, 6)<<0;	//b1[5:0],
		rgb[2][2] |= ReadBits(in, &bit, 4)<<0;	//b2[3:0],
		rgb[2][0] |= ReadBits(in, &bit, 6)<<0;	//r2[5:0],
		rgb[3][0] |= ReadBits(in, &bit, 6)<<0;	//r3[5:0]
		shapeindex = ReadBits(in, &bit, 5);
		break;
	case 2:	//3
		Vector4Set(cbits, 5,4,4, 11);
		tr = 1;
		ss = 2;
		rgb[0][0] |= ReadBitsL(in, &bit,10)<<0;	//r0[9:0],
		rgb[0][1] |= ReadBitsL(in, &bit,10)<<0;	//g0[9:0],
		rgb[0][2] |= ReadBitsL(in, &bit,10)<<0;	//b0[9:0],
		rgb[1][0] |= ReadBits(in, &bit, 5)<<0;	//r1[4:0],
		rgb[0][0] |= ReadBits(in, &bit, 1)<<10;	//r0[10],
		rgb[2][1] |= ReadBits(in, &bit, 4)<<0;	//g2[3:0],
		rgb[1][1] |= ReadBits(in, &bit, 4)<<0;	//g1[3:0],
		rgb[0][1] |= ReadBits(in, &bit, 1)<<10;	//g0[10],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<0;	//b3[0],
		rgb[3][1] |= ReadBits(in, &bit, 4)<<0;	//g3[3:0],
		rgb[1][2] |= ReadBits(in, &bit, 4)<<0;	//b1[3:0],
		rgb[0][2] |= ReadBits(in, &bit, 1)<<10;	//b0[10],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<1;	//b3[1],
		rgb[2][2] |= ReadBits(in, &bit, 4)<<0;	//b2[3:0],
		rgb[2][0] |= ReadBits(in, &bit, 5)<<0;	//r2[4:0],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<2;	//b3[2],
		rgb[3][0] |= ReadBits(in, &bit, 5)<<0;	//r3[4:0],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<3;	//b3[3]
		shapeindex = ReadBits(in, &bit, 5);
		break;
	case 6:	//4
		Vector4Set(cbits, 4,5,4, 11);
		tr = 1;
		ss = 2;
		rgb[0][0] |= ReadBitsL(in, &bit,10)<<0;	//r0[9:0],
		rgb[0][1] |= ReadBitsL(in, &bit,10)<<0;	//g0[9:0],
		rgb[0][2] |= ReadBitsL(in, &bit,10)<<0;	//b0[9:0],
		rgb[1][0] |= ReadBits(in, &bit, 4)<<0;	//r1[3:0],
		rgb[0][0] |= ReadBits(in, &bit, 1)<<10;	//r0[10],
		rgb[3][1] |= ReadBits(in, &bit, 1)<<4;	//g3[4],
		rgb[2][1] |= ReadBits(in, &bit, 4)<<0;	//g2[3:0],
		rgb[1][1] |= ReadBits(in, &bit, 5)<<0;	//g1[4:0],
		rgb[0][1] |= ReadBits(in, &bit, 1)<<10;	//g0[10],
		rgb[3][1] |= ReadBits(in, &bit, 4)<<0;	//g3[3:0],
		rgb[1][2] |= ReadBits(in, &bit, 4)<<0;	//b1[3:0],
		rgb[0][2] |= ReadBits(in, &bit, 1)<<10;	//b0[10],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<1;	//b3[1],
		rgb[2][2] |= ReadBits(in, &bit, 4)<<0;	//b2[3:0],
		rgb[2][0] |= ReadBits(in, &bit, 4)<<0;	//r2[3:0],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<0;	//b3[0],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<2;	//b3[2],
		rgb[3][0] |= ReadBits(in, &bit, 4)<<0;	//r3[3:0],
		rgb[2][1] |= ReadBits(in, &bit, 1)<<4;	//g2[4],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<3;	//b3[3]
		shapeindex = ReadBits(in, &bit, 5);
		break;
	case 10: //5
		Vector4Set(cbits, 4,4,5, 11);
		tr = 1;
		ss = 2;
		rgb[0][0] |= ReadBitsL(in, &bit,10)<<0;	//r0[9:0],
		rgb[0][1] |= ReadBitsL(in, &bit,10)<<0;	//g0[9:0],
		rgb[0][2] |= ReadBitsL(in, &bit,10)<<0;	//b0[9:0],
		rgb[1][0] |= ReadBits(in, &bit, 4)<<0;	//r1[3:0],
		rgb[0][0] |= ReadBits(in, &bit, 1)<<10;	//r0[10],
		rgb[2][2] |= ReadBits(in, &bit, 1)<<4;	//b2[4],
		rgb[2][1] |= ReadBits(in, &bit, 4)<<0;	//g2[3:0],
		rgb[1][1] |= ReadBits(in, &bit, 4)<<0;	//g1[3:0],
		rgb[0][1] |= ReadBits(in, &bit, 1)<<10;	//g0[10],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<0;	//b3[0],
		rgb[3][1] |= ReadBits(in, &bit, 4)<<0;	//g3[3:0],
		rgb[1][2] |= ReadBits(in, &bit, 5)<<0;	//b1[4:0],
		rgb[0][2] |= ReadBits(in, &bit, 1)<<10;	//b0[10],
		rgb[2][2] |= ReadBits(in, &bit, 4)<<0;	//b2[3:0],
		rgb[2][0] |= ReadBits(in, &bit, 4)<<0;	//r2[3:0],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<1;	//b3[1],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<2;	//b3[2],
		rgb[3][0] |= ReadBits(in, &bit, 4)<<0;	//r3[3:0],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<4;	//b3[4],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<3;	//b3[3]
		shapeindex = ReadBits(in, &bit, 5);
		break;
	case 14:	//6
		Vector4Set(cbits, 5,5,5, 9);
		tr = 1;
		ss = 2;
		rgb[0][0] |= ReadBits(in, &bit, 9)<<0;	//r0[8:0],
		rgb[2][2] |= ReadBits(in, &bit, 1)<<4;	//b2[4],
		rgb[0][1] |= ReadBits(in, &bit, 9)<<0;	//g0[8:0],
		rgb[2][1] |= ReadBits(in, &bit, 1)<<4;	//g2[4],
		rgb[0][2] |= ReadBits(in, &bit, 9)<<0;	//b0[8:0],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<4;	//b3[4],
		rgb[1][0] |= ReadBits(in, &bit, 5)<<0;	//r1[4:0],
		rgb[3][1] |= ReadBits(in, &bit, 1)<<4;	//g3[4],
		rgb[2][1] |= ReadBits(in, &bit, 4)<<0;	//g2[3:0],
		rgb[1][1] |= ReadBits(in, &bit, 5)<<0;	//g1[4:0],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<0;	//b3[0],
		rgb[3][1] |= ReadBits(in, &bit, 4)<<0;	//g3[3:0],
		rgb[1][2] |= ReadBits(in, &bit, 5)<<0;	//b1[4:0],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<1;	//b3[1],
		rgb[2][2] |= ReadBits(in, &bit, 4)<<0;	//b2[3:0],
		rgb[2][0] |= ReadBits(in, &bit, 5)<<0;	//r2[4:0],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<2;	//b3[2],
		rgb[3][0] |= ReadBits(in, &bit, 5)<<0;	//r3[4:0],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<3;	//b3[3]
		shapeindex = ReadBits(in, &bit, 5);
		break;
	case 18:	//7
		Vector4Set(cbits, 6,5,5, 8);
		tr = 1;
		ss = 2;
		rgb[0][0] |= ReadBits(in, &bit, 8)<<0;	//r0[7:0],
		rgb[3][1] |= ReadBits(in, &bit, 1)<<4;	//g3[4],
		rgb[2][2] |= ReadBits(in, &bit, 1)<<4;	//b2[4],
		rgb[0][1] |= ReadBits(in, &bit, 8)<<0;	//g0[7:0],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<2;	//b3[2],
		rgb[2][1] |= ReadBits(in, &bit, 1)<<4;	//g2[4],
		rgb[0][2] |= ReadBits(in, &bit, 8)<<0;	//b0[7:0],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<3;	//b3[3],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<4;	//b3[4],
		rgb[1][0] |= ReadBits(in, &bit, 6)<<0;	//r1[5:0],
		rgb[2][1] |= ReadBits(in, &bit, 4)<<0;	//g2[3:0],
		rgb[1][1] |= ReadBits(in, &bit, 5)<<0;	//g1[4:0],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<0;	//b3[0],
		rgb[3][1] |= ReadBits(in, &bit, 4)<<0;	//g3[3:0],
		rgb[1][2] |= ReadBits(in, &bit, 5)<<0;	//b1[4:0],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<1;	//b3[1],
		rgb[2][2] |= ReadBits(in, &bit, 4)<<0;	//b2[3:0],
		rgb[2][0] |= ReadBits(in, &bit, 6)<<0;	//r2[5:0],
		rgb[3][0] |= ReadBits(in, &bit, 6)<<0;	//r3[5:0]
		shapeindex = ReadBits(in, &bit, 5);
		break;
	case 22:	//8
		Vector4Set(cbits, 5,6,5, 8);
		tr = 1;
		ss = 2;
		rgb[0][0] |= ReadBits(in, &bit, 8)<<0;	//r0[7:0],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<0;	//b3[0],
		rgb[2][2] |= ReadBits(in, &bit, 1)<<4;	//b2[4],
		rgb[0][1] |= ReadBits(in, &bit, 8)<<0;	//g0[7:0],
		rgb[2][1] |= ReadBits(in, &bit, 1)<<5;	//g2[5],
		rgb[2][1] |= ReadBits(in, &bit, 1)<<4;	//g2[4],
		rgb[0][2] |= ReadBits(in, &bit, 8)<<0;	//b0[7:0],
		rgb[3][1] |= ReadBits(in, &bit, 1)<<5;	//g3[5],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<4;	//b3[4],
		rgb[1][0] |= ReadBits(in, &bit, 5)<<0;	//r1[4:0],
		rgb[3][1] |= ReadBits(in, &bit, 1)<<4;	//g3[4],
		rgb[2][1] |= ReadBits(in, &bit, 4)<<0;	//g2[3:0],
		rgb[1][1] |= ReadBits(in, &bit, 6)<<0;	//g1[5:0],
		rgb[3][1] |= ReadBits(in, &bit, 4)<<0;	//g3[3:0],
		rgb[1][2] |= ReadBits(in, &bit, 5)<<0;	//b1[4:0],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<1;	//b3[1],
		rgb[2][2] |= ReadBits(in, &bit, 4)<<0;	//b2[3:0],
		rgb[2][0] |= ReadBits(in, &bit, 5)<<0;	//r2[4:0],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<2;	//b3[2],
		rgb[3][0] |= ReadBits(in, &bit, 5)<<0;	//r3[4:0],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<3;	//b3[3]
		shapeindex = ReadBits(in, &bit, 5);
		break;
	case 26:	//9
		Vector4Set(cbits, 5,5,6, 8);
		tr = 1;
		ss = 2;
		rgb[0][0] |= ReadBits(in, &bit, 8)<<0;	//r0[7:0],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<1;	//b3[1],
		rgb[2][2] |= ReadBits(in, &bit, 1)<<4;	//b2[4],
		rgb[0][1] |= ReadBits(in, &bit, 8)<<0;	//g0[7:0],
		rgb[2][2] |= ReadBits(in, &bit, 1)<<5;	//b2[5],
		rgb[2][1] |= ReadBits(in, &bit, 1)<<4;	//g2[4],
		rgb[0][2] |= ReadBits(in, &bit, 8)<<0;	//b0[7:0],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<5;	//b3[5],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<4;	//b3[4],
		rgb[1][0] |= ReadBits(in, &bit, 5)<<0;	//r1[4:0],
		rgb[3][1] |= ReadBits(in, &bit, 1)<<4;	//g3[4],
		rgb[2][1] |= ReadBits(in, &bit, 4)<<0;	//g2[3:0],
		rgb[1][1] |= ReadBits(in, &bit, 5)<<0;	//g1[4:0],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<0;	//b3[0],
		rgb[3][1] |= ReadBits(in, &bit, 4)<<0;	//g3[3:0],
		rgb[1][2] |= ReadBits(in, &bit, 6)<<0;	//b1[5:0],
		rgb[2][2] |= ReadBits(in, &bit, 4)<<0;	//b2[3:0],
		rgb[2][0] |= ReadBits(in, &bit, 5)<<0;	//r2[4:0],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<2;	//b3[2],
		rgb[3][0] |= ReadBits(in, &bit, 5)<<0;	//r3[4:0],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<3;	//b3[3]
		shapeindex = ReadBits(in, &bit, 5);
		break;
	case 30:	//10
		Vector4Set(cbits, 6,6,6, 6);
		tr = 0;
		ss = 2;
		rgb[0][0] |= ReadBits(in, &bit, 6)<<0;	//r0[5:0],
		rgb[3][1] |= ReadBits(in, &bit, 1)<<4;	//g3[4],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<0;	//b3[0],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<1;	//b3[1],
		rgb[2][2] |= ReadBits(in, &bit, 1)<<4;	//b2[4],
		rgb[0][1] |= ReadBits(in, &bit, 6)<<0;	//g0[5:0],
		rgb[2][1] |= ReadBits(in, &bit, 1)<<5;	//g2[5],
		rgb[2][2] |= ReadBits(in, &bit, 1)<<5;	//b2[5],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<2;	//b3[2],
		rgb[2][1] |= ReadBits(in, &bit, 1)<<4;	//g2[4],
		rgb[0][2] |= ReadBits(in, &bit, 6)<<0;	//b0[5:0],
		rgb[3][1] |= ReadBits(in, &bit, 1)<<5;	//g3[5],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<3;	//b3[3],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<5;	//b3[5],
		rgb[3][2] |= ReadBits(in, &bit, 1)<<4;	//b3[4],
		rgb[1][0] |= ReadBits(in, &bit, 6)<<0;	//r1[5:0],
		rgb[2][1] |= ReadBits(in, &bit, 4)<<0;	//g2[3:0],
		rgb[1][1] |= ReadBits(in, &bit, 6)<<0;	//g1[5:0],
		rgb[3][1] |= ReadBits(in, &bit, 4)<<0;	//g3[3:0],
		rgb[1][2] |= ReadBits(in, &bit, 6)<<0;	//b1[5:0],
		rgb[2][2] |= ReadBits(in, &bit, 4)<<0;	//b2[3:0],
		rgb[2][0] |= ReadBits(in, &bit, 6)<<0;	//r2[5:0],
		rgb[3][0] |= ReadBits(in, &bit, 6)<<0;	//r3[5:0]
		shapeindex = ReadBits(in, &bit, 5);
		break;
	case 3: //11
		Vector4Set(cbits, 10,10,10, 10);
		tr = 0;
		ss = 1;
		rgb[0][0] |= ReadBitsL(in, &bit,10)<<0;	//r0[9:0]
		rgb[0][1] |= ReadBitsL(in, &bit,10)<<0;	//g0[9:0]
		rgb[0][2] |= ReadBitsL(in, &bit,10)<<0;	//b0[9:0]
		rgb[1][0] |= ReadBitsL(in, &bit,10)<<0;	//r1[9:0]
		rgb[1][1] |= ReadBitsL(in, &bit,10)<<0;	//g1[9:0]
		rgb[1][2] |= ReadBitsL(in, &bit,10)<<0;	//b1[9:0]
		shapeindex = 0;
		break;
	case 7: //12
		Vector4Set(cbits, 9,9,9, 11);
		tr = 1;
		ss = 1;
		rgb[0][0] |= ReadBitsL(in, &bit,10)<<0;	//r0[9:0],
		rgb[0][1] |= ReadBitsL(in, &bit,10)<<0;	//g0[9:0],
		rgb[0][2] |= ReadBitsL(in, &bit,10)<<0;	//b0[9:0],
		rgb[1][0] |= ReadBits(in, &bit, 9)<<0;	//r1[8:0],
		rgb[0][0] |= ReadBits(in, &bit, 1)<<10;	//r0[10],
		rgb[1][1] |= ReadBits(in, &bit, 9)<<0;	//g1[8:0],
		rgb[0][1] |= ReadBits(in, &bit, 1)<<10;	//g0[10],
		rgb[1][2] |= ReadBits(in, &bit, 9)<<0;	//b1[8:0],
		rgb[0][2] |= ReadBits(in, &bit, 1)<<10;	//b0[10]
		shapeindex = 0;
		break;
	case 11: //13
		Vector4Set(cbits, 8,8,8, 12);
		tr = 1;
		ss = 1;
		rgb[0][0] |= ReadBitsL(in, &bit,10)<<0;	//r0[9:0],
		rgb[0][1] |= ReadBitsL(in, &bit,10)<<0;	//g0[9:0],
		rgb[0][2] |= ReadBitsL(in, &bit,10)<<0;	//b0[9:0],
		rgb[1][0] |= ReadBits(in, &bit, 8)<<0;	//r1[7:0],
		rgb[0][0] |= ReadBits(in, &bit, 1)<<11;	//r0[11],
		rgb[0][0] |= ReadBits(in, &bit, 1)<<10;	//r0[10],
		rgb[1][1] |= ReadBits(in, &bit, 8)<<0;	//g1[7:0],
		rgb[0][1] |= ReadBits(in, &bit, 1)<<11;	//g0[11],
		rgb[0][1] |= ReadBits(in, &bit, 1)<<10;	//g0[10],
		rgb[1][2] |= ReadBits(in, &bit, 8)<<0;	//b1[7:0],
		rgb[0][2] |= ReadBits(in, &bit, 1)<<11;	//b0[11],
		rgb[0][2] |= ReadBits(in, &bit, 1)<<10;	//b0[10]
		shapeindex = 0;
		break;
	case 15: //14
		//UNTESTED
		Vector4Set(cbits, 4,4,4, 16);
		tr = 1;
		ss = 1;
		rgb[0][0] |= ReadBitsL(in, &bit,10)<<0;	//r0[9:0],
		rgb[0][1] |= ReadBitsL(in, &bit,10)<<0;	//g0[9:0],
		rgb[0][2] |= ReadBitsL(in, &bit,10)<<0;	//b0[9:0],
		rgb[1][0] |= ReadBits(in, &bit, 4)<<0;	//r1[3:0],
		rgb[0][0] |= ReadBits(in, &bit, 6)<<10;	//r0[10:15],
		rgb[1][1] |= ReadBits(in, &bit, 4)<<0;	//g1[3:0],
		rgb[0][1] |= ReadBits(in, &bit, 6)<<10;	//g0[10:15],
		rgb[1][2] |= ReadBits(in, &bit, 4)<<0;	//b1[3:0],
		rgb[0][2] |= ReadBits(in, &bit, 6)<<10;	//b0[10:15]
		shapeindex = 0;
		break;

	default:
		//reserved modes
		for (i = 0; i < 16; )
		{
			Vector4Set(out[i].v, 0, 0, 0, 0xf<<10);
			i++;
			if (!(i & 3))
				out += w-4;
		}
		return;
	}

	ib = (ss==2)?3:4;
	cb = 1<<ib;
	weight = wsz[ib];
	if (signextend)
	{
		int v[2], q, k, g, s;
		for (g = 0; g < ss; g++)	//subsets
		{
			for (j = 0; j < 3; j++)	//channels
			{
				for (k = 0; k < 2; k++)	//start/end
				{
					v[k] = rgb[k+g*2][j];
					if ((k || g) && tr)
					{	//the specified colour values are deltas from the first colour value
						if (v[k] & (1<<(cbits[j]-1)))	//high bit set
							v[k] |= ~0<<cbits[j];		//sign extend it...
						v[k] = (int)rgb[0][j] + v[k];		//add it to the base
						v[k] &= ((1<<cbits[3])-1);		//and mask it to avoid overflows...
					}
					if (v[k] & (1<<(cbits[3]-1)))	//high bit set
						v[k] |= ~0<<cbits[3];		//sign extend it

					if (cbits[3] >= 16)
						;
					else
					{
						s = v[k] < 0;
						if (s)
							v[k]=-v[k];
						if (v[k] == 0)
							v[k] = 0;
						else if (v[k] >= (1<<(cbits[3]-1))-1)
							v[k] = 0x7fff;
						else
							v[k] = (v[k] * (0x7fff+1) + (0x7fff+1)/2) >> (cbits[3]-1);
						if (s)
							v[k] = -v[k];
					}
				}
				for (i = 0; i < cb; i++)	//indexbits
				{
					q = (v[0]*(64-weight[i]) + v[1]*weight[i])>>6;	//this ignores sign, which seems somewhat dodgy.
					q = (q < 0) ? -(((-q) * 31) >> 5) : (q * 31) >> 5;
					palette[i+g*8][j] = q;
				}
			}
		}
	}
	else
	{
		int v[2], q, k, g;
		for (g = 0; g < ss; g++)	//subsets
		{
			for (j = 0; j < 3; j++)	//channels
			{
				for (k = 0; k < 2; k++)	//start/end
				{
					v[k] = rgb[k+g*2][j];
					if ((k || g) && tr)
					{	//the specified colour values are deltas from the first colour value
						if (v[k] & (1<<(cbits[j]-1)))	//high bit set
							v[k] |= ~0<<cbits[j];		//sign extend it...
						v[k] = (int)rgb[0][j] + v[k];		//add it to the base
						v[k] &= ((1<<cbits[3])-1);		//and mask it to avoid overflows...
//						if (v[k] & (1<<(bits-1)))	//high bit set
//							v[k] |= ~0<<bits;		//sign extend it
					}

					if (cbits[3] >= 15)
						;
					else if (v[k] == 0)
						v[k] = 0;
					else if (v[k] == (1<<cbits[3])-1)
						v[k] = 0xffff;
					else
						v[k] = (v[k] * (0xffff+1) + (0xffff+1)/2) >> cbits[3];
				}
				for (i = 0; i < cb; i++)	//indexbits
				{
					q = (v[0]*(64-weight[i]) + v[1]*weight[i])>>6;
					q = (q*31)>>6;
					palette[i+g*8][j] = q;
				}
			}
		}
	}

	if (ss == 2)
	{
		const qbyte *p = p2+shapeindex*16;
		for (i = 0; i < 16; )
		{
			int pidx = p[i];
			int idx;
			if (i == 0 || i == anchors[shapeindex])
				idx = ReadBits(in, &bit, ib-1);
			else
				idx = ReadBits(in, &bit, ib);
			if (pidx)
				idx += 8;
			out[i].v[0] = palette[idx][0];
			out[i].v[1] = palette[idx][1];
			out[i].v[2] = palette[idx][2];
			out[i].v[3] = 0xf<<10;	//must be 1
			i++;
			if (!(i & 3))
				out += w-4;
		}
	}
	else
	{
		for (i = 0; i < 16; )
		{
	//		int pidx = p[i];
			int idx;
			if (i == 0)
				idx = ReadBits(in, &bit, ib-1);
			else
				idx = ReadBits(in, &bit, ib);
			out[i].v[0] = palette[idx][0];
			out[i].v[1] = palette[idx][1];
			out[i].v[2] = palette[idx][2];
			out[i].v[3] = 0xf<<10;	//must be 1
			i++;
			if (!(i & 3))
				out += w-4;
		}
	}
	if (bit != 128)
	{
		out -= 4*(w-4);	//undo the above damage
//		out->v[0] = 0;//0xf<<10;	//must be 1
//		out->v[1] = 0;//0xf<<10;	//must be 1
//		out->v[2] = 0;//0xf<<10;	//must be 1
		out->v[3] = 0;//0xf<<10;	//must be 1
	}
}
static void Image_Decode_BC6S_Block(qbyte *fte_restrict in, pixel64_t *fte_restrict out, int w)
{
	Image_Decode_BC6_Block(in,out,w,true);
}
static void Image_Decode_BC6U_Block(qbyte *fte_restrict in, pixel64_t *fte_restrict out, int w)
{
	Image_Decode_BC6_Block(in,out,w,false);
}
static void Image_Decode_BC7_Block(qbyte *fte_restrict in, pixel32_t *fte_restrict out, int w)
{
	static const qbyte p1[] = {
		0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,
	};
	static const qbyte p2[] = {
		0,0,1,1,0,0,1,1,0,0,1,1,0,0,1,1,
		0,0,0,1,0,0,0,1,0,0,0,1,0,0,0,1,
		0,1,1,1,0,1,1,1,0,1,1,1,0,1,1,1,
		0,0,0,1,0,0,1,1,0,0,1,1,0,1,1,1,
		0,0,0,0,0,0,0,1,0,0,0,1,0,0,1,1,
		0,0,1,1,0,1,1,1,0,1,1,1,1,1,1,1,
		0,0,0,1,0,0,1,1,0,1,1,1,1,1,1,1,
		0,0,0,0,0,0,0,1,0,0,1,1,0,1,1,1,
		0,0,0,0,0,0,0,0,0,0,0,1,0,0,1,1,
		0,0,1,1,0,1,1,1,1,1,1,1,1,1,1,1,
		0,0,0,0,0,0,0,1,0,1,1,1,1,1,1,1,
		0,0,0,0,0,0,0,0,0,0,0,1,0,1,1,1,
		0,0,0,1,0,1,1,1,1,1,1,1,1,1,1,1,
		0,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,
		0,0,0,0,1,1,1,1,1,1,1,1,1,1,1,1,
		0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,
		0,0,0,0,1,0,0,0,1,1,1,0,1,1,1,1,
		0,1,1,1,0,0,0,1,0,0,0,0,0,0,0,0,
		0,0,0,0,0,0,0,0,1,0,0,0,1,1,1,0,
		0,1,1,1,0,0,1,1,0,0,0,1,0,0,0,0,
		0,0,1,1,0,0,0,1,0,0,0,0,0,0,0,0,
		0,0,0,0,1,0,0,0,1,1,0,0,1,1,1,0,
		0,0,0,0,0,0,0,0,1,0,0,0,1,1,0,0,
		0,1,1,1,0,0,1,1,0,0,1,1,0,0,0,1,
		0,0,1,1,0,0,0,1,0,0,0,1,0,0,0,0,
		0,0,0,0,1,0,0,0,1,0,0,0,1,1,0,0,
		0,1,1,0,0,1,1,0,0,1,1,0,0,1,1,0,
		0,0,1,1,0,1,1,0,0,1,1,0,1,1,0,0,
		0,0,0,1,0,1,1,1,1,1,1,0,1,0,0,0,
		0,0,0,0,1,1,1,1,1,1,1,1,0,0,0,0,
		0,1,1,1,0,0,0,1,1,0,0,0,1,1,1,0,
		0,0,1,1,1,0,0,1,1,0,0,1,1,1,0,0,
		0,1,0,1,0,1,0,1,0,1,0,1,0,1,0,1,
		0,0,0,0,1,1,1,1,0,0,0,0,1,1,1,1,
		0,1,0,1,1,0,1,0,0,1,0,1,1,0,1,0,
		0,0,1,1,0,0,1,1,1,1,0,0,1,1,0,0,
		0,0,1,1,1,1,0,0,0,0,1,1,1,1,0,0,
		0,1,0,1,0,1,0,1,1,0,1,0,1,0,1,0,
		0,1,1,0,1,0,0,1,0,1,1,0,1,0,0,1,
		0,1,0,1,1,0,1,0,1,0,1,0,0,1,0,1,
		0,1,1,1,0,0,1,1,1,1,0,0,1,1,1,0,
		0,0,0,1,0,0,1,1,1,1,0,0,1,0,0,0,
		0,0,1,1,0,0,1,0,0,1,0,0,1,1,0,0,
		0,0,1,1,1,0,1,1,1,1,0,1,1,1,0,0,
		0,1,1,0,1,0,0,1,1,0,0,1,0,1,1,0,
		0,0,1,1,1,1,0,0,1,1,0,0,0,0,1,1,
		0,1,1,0,0,1,1,0,1,0,0,1,1,0,0,1,
		0,0,0,0,0,1,1,0,0,1,1,0,0,0,0,0,
		0,1,0,0,1,1,1,0,0,1,0,0,0,0,0,0,
		0,0,1,0,0,1,1,1,0,0,1,0,0,0,0,0,
		0,0,0,0,0,0,1,0,0,1,1,1,0,0,1,0,
		0,0,0,0,0,1,0,0,1,1,1,0,0,1,0,0,
		0,1,1,0,1,1,0,0,1,0,0,1,0,0,1,1,
		0,0,1,1,0,1,1,0,1,1,0,0,1,0,0,1,
		0,1,1,0,0,0,1,1,1,0,0,1,1,1,0,0,
		0,0,1,1,1,0,0,1,1,1,0,0,0,1,1,0,
		0,1,1,0,1,1,0,0,1,1,0,0,1,0,0,1,
		0,1,1,0,0,0,1,1,0,0,1,1,1,0,0,1,
		0,1,1,1,1,1,1,0,1,0,0,0,0,0,0,1,
		0,0,0,1,1,0,0,0,1,1,1,0,0,1,1,1,
		0,0,0,0,1,1,1,1,0,0,1,1,0,0,1,1,
		0,0,1,1,0,0,1,1,1,1,1,1,0,0,0,0,
		0,0,1,0,0,0,1,0,1,1,1,0,1,1,1,0,
		0,1,0,0,0,1,0,0,0,1,1,1,0,1,1,1,
    };
    static const qbyte p3[] = {
		0,0,1,1,0,0,1,1,0,2,2,1,2,2,2,2,
		0,0,0,1,0,0,1,1,2,2,1,1,2,2,2,1,
		0,0,0,0,2,0,0,1,2,2,1,1,2,2,1,1,
		0,2,2,2,0,0,2,2,0,0,1,1,0,1,1,1,
		0,0,0,0,0,0,0,0,1,1,2,2,1,1,2,2,
		0,0,1,1,0,0,1,1,0,0,2,2,0,0,2,2,
		0,0,2,2,0,0,2,2,1,1,1,1,1,1,1,1,
		0,0,1,1,0,0,1,1,2,2,1,1,2,2,1,1,
		0,0,0,0,0,0,0,0,1,1,1,1,2,2,2,2,
		0,0,0,0,1,1,1,1,1,1,1,1,2,2,2,2,
		0,0,0,0,1,1,1,1,2,2,2,2,2,2,2,2,
		0,0,1,2,0,0,1,2,0,0,1,2,0,0,1,2,
		0,1,1,2,0,1,1,2,0,1,1,2,0,1,1,2,
		0,1,2,2,0,1,2,2,0,1,2,2,0,1,2,2,
		0,0,1,1,0,1,1,2,1,1,2,2,1,2,2,2,
		0,0,1,1,2,0,0,1,2,2,0,0,2,2,2,0,
		0,0,0,1,0,0,1,1,0,1,1,2,1,1,2,2,
		0,1,1,1,0,0,1,1,2,0,0,1,2,2,0,0,
		0,0,0,0,1,1,2,2,1,1,2,2,1,1,2,2,
		0,0,2,2,0,0,2,2,0,0,2,2,1,1,1,1,
		0,1,1,1,0,1,1,1,0,2,2,2,0,2,2,2,
		0,0,0,1,0,0,0,1,2,2,2,1,2,2,2,1,
		0,0,0,0,0,0,1,1,0,1,2,2,0,1,2,2,
		0,0,0,0,1,1,0,0,2,2,1,0,2,2,1,0,
		0,1,2,2,0,1,2,2,0,0,1,1,0,0,0,0,
		0,0,1,2,0,0,1,2,1,1,2,2,2,2,2,2,
		0,1,1,0,1,2,2,1,1,2,2,1,0,1,1,0,
		0,0,0,0,0,1,1,0,1,2,2,1,1,2,2,1,
		0,0,2,2,1,1,0,2,1,1,0,2,0,0,2,2,
		0,1,1,0,0,1,1,0,2,0,0,2,2,2,2,2,
		0,0,1,1,0,1,2,2,0,1,2,2,0,0,1,1,
		0,0,0,0,2,0,0,0,2,2,1,1,2,2,2,1,
		0,0,0,0,0,0,0,2,1,1,2,2,1,2,2,2,
		0,2,2,2,0,0,2,2,0,0,1,2,0,0,1,1,
		0,0,1,1,0,0,1,2,0,0,2,2,0,2,2,2,
		0,1,2,0,0,1,2,0,0,1,2,0,0,1,2,0,
		0,0,0,0,1,1,1,1,2,2,2,2,0,0,0,0,
		0,1,2,0,1,2,0,1,2,0,1,2,0,1,2,0,
		0,1,2,0,2,0,1,2,1,2,0,1,0,1,2,0,
		0,0,1,1,2,2,0,0,1,1,2,2,0,0,1,1,
		0,0,1,1,1,1,2,2,2,2,0,0,0,0,1,1,
		0,1,0,1,0,1,0,1,2,2,2,2,2,2,2,2,
		0,0,0,0,0,0,0,0,2,1,2,1,2,1,2,1,
		0,0,2,2,1,1,2,2,0,0,2,2,1,1,2,2,
		0,0,2,2,0,0,1,1,0,0,2,2,0,0,1,1,
		0,2,2,0,1,2,2,1,0,2,2,0,1,2,2,1,
		0,1,0,1,2,2,2,2,2,2,2,2,0,1,0,1,
		0,0,0,0,2,1,2,1,2,1,2,1,2,1,2,1,
		0,1,0,1,0,1,0,1,0,1,0,1,2,2,2,2,
		0,2,2,2,0,1,1,1,0,2,2,2,0,1,1,1,
		0,0,0,2,1,1,1,2,0,0,0,2,1,1,1,2,
		0,0,0,0,2,1,1,2,2,1,1,2,2,1,1,2,
		0,2,2,2,0,1,1,1,0,1,1,1,0,2,2,2,
		0,0,0,2,1,1,1,2,1,1,1,2,0,0,0,2,
		0,1,1,0,0,1,1,0,0,1,1,0,2,2,2,2,
		0,0,0,0,0,0,0,0,2,1,1,2,2,1,1,2,
		0,1,1,0,0,1,1,0,2,2,2,2,2,2,2,2,
		0,0,2,2,0,0,1,1,0,0,1,1,0,0,2,2,
		0,0,2,2,1,1,2,2,1,1,2,2,0,0,2,2,
		0,0,0,0,0,0,0,0,0,0,0,0,2,1,1,2,
		0,0,0,2,0,0,0,1,0,0,0,2,0,0,0,1,
		0,2,2,2,1,2,2,2,0,2,2,2,1,2,2,2,
		0,1,0,1,2,2,2,2,2,2,2,2,2,2,2,2,
		0,1,1,1,2,0,1,1,2,2,0,1,2,2,2,0,
    };

    static const qbyte *psz[] = {p1,p1,p2,p3};

	static const qbyte anchortable[][64] = {{
		15,15,15,15,15,15,15,15,
		15,15,15,15,15,15,15,15,
		15, 2, 8, 2, 2, 8, 8,15,
		 2, 8, 2, 2, 8, 8, 2, 2,
		15,15, 6, 8, 2, 8,15,15,
		 2, 8, 2, 2, 2,15,15, 6,
		 6, 2, 6, 8,15,15, 2, 2,
		15,15,15,15,15, 2, 2,15,
    },{
		 3, 3,15,15, 8, 3,15,15,
		 8, 8, 6, 6, 6, 5, 3, 3,
		 3, 3, 8,15, 3, 3, 6,10,
		 5, 8, 8, 6, 8, 5,15,15,
		 8,15, 3, 5, 6,10, 8,15,
		15, 3,15, 5,15,15,15,15,
		 3,15, 5, 5, 5, 8, 5,10,
		 5,10, 8,13,15,12, 3, 3,
	},{
		15, 8, 8, 3,15,15, 3, 8,
		15,15,15,15,15,15,15, 8,
		15, 8,15, 3,15, 8,15, 8,
		 3,15, 6,10,15,15,10, 8,
		15, 3,15,10,10, 8, 9,10,
		 6,15, 8,15, 3, 6, 6, 8,
		15, 3,15,15,15,15,15,15,
		15,15,15,15, 3,15,15, 8,
    }};
	int anchor[3];

	static const int w1[] = {0, 64};
    static const int w2[] = {0, 21, 43, 64};
    static const int w3[] = {0, 9, 18, 27, 37, 46, 55, 64};
    static const int w4[] = {0, 4, 9, 13, 17, 21, 26, 30, 34, 38, 43, 47, 51, 55, 60, 64};
    static const int *wsz[] = {w1,w1, w2,w3,w4};
    static const struct {
		int numsubsets;
		int partitionbits;
		int rotationbits;
		int indexselectionbits;
		int colourbits;
		int alphabits;
		int pmode;
		int indexbits[2];
    } m[9] =
    {
		{3, 4, 0, 0, 4, 0, 1, {3, 0}},	//mode 0 - 3*rgb4
		{2, 6, 0, 0, 6, 0, 2, {3, 0}},	//mode 1 - 2*rgb6
		{3, 6, 0, 0, 5, 0, 0, {2, 0}},	//mode 2 - 3*rgb5
		{2, 6, 0, 0, 7, 0, 1, {2, 0}},	//mode 3 - 2*rgb7
		{1, 0, 2, 1, 5, 6, 0, {2, 3}},	//mode 4 - 1*rgb5+a6
		{1, 0, 2, 0, 7, 8, 0, {2, 2}},	//mode 5 - 1*rgb7+a8
		{1, 0, 0, 0, 7, 7, 1, {4, 0}},	//mode 6 - 1*rgb7a7
		{2, 6, 0, 0, 5, 5, 1, {2, 0}},	//mode 7 - 2*rgb5a5
		{1, 0, 0, 0, 0, 0, 0, {0, 0}},	//mode 8 - reserved
    };

	pixel32_t palette[3][2];
	pixel32_t tab[3][16];
	int mode, i, j, bit, partition, ss, cb;
	const int *weight;
	const qbyte *p;
	int rot;
	int idxsel;
	for (mode = 0; mode < 8; mode++)
		if (*in & (1u<<mode))
			break;
	ss = m[mode].numsubsets;
	bit = mode+1;
	partition = ReadBits(in, &bit, m[mode].partitionbits);
	rot = ReadBits(in, &bit, m[mode].rotationbits);
	idxsel = ReadBits(in, &bit, m[mode].indexselectionbits);
	p = psz[ss] + partition*16;
	for (j = 0; j < 3; j++)
		for (i = 0; i < ss; i++)
		{
			palette[i][0].v[j] = ReadBits(in, &bit, m[mode].colourbits) << (8-m[mode].colourbits);
			palette[i][1].v[j] = ReadBits(in, &bit, m[mode].colourbits) << (8-m[mode].colourbits);
		}
	for (i = 0; i < ss; i++)
	{
		if (m[mode].alphabits)
		{
			palette[i][0].v[3] = ReadBits(in, &bit, m[mode].alphabits) << (8-m[mode].alphabits);
			palette[i][1].v[3] = ReadBits(in, &bit, m[mode].alphabits) << (8-m[mode].alphabits);
		}
		else palette[i][0].v[3] = palette[i][1].v[3] = 255;
	}

	if(m[mode].pmode)
	{
		for (i = 0; i < m[mode].numsubsets; i++)
		{
			qbyte p = ReadBits(in, &bit, 1);
			for (j = 0; j < 3; j++)
				palette[i][0].v[j] |= p<<(7-m[mode].colourbits);
			palette[i][0].v[3] |= p<<(7-m[mode].alphabits);

			if (m[mode].pmode!=2)
				p = ReadBits(in, &bit, 1);
			for (j = 0; j < 3; j++)
				palette[i][1].v[j] |= p<<(7-m[mode].colourbits);
			palette[i][1].v[3] |= p<<(7-m[mode].alphabits);
		}
		etc_expandv(palette[i][0], m[mode].colourbits+1, m[mode].colourbits+1, m[mode].colourbits+1); palette[i][0].v[3]|=palette[i][0].v[3]>>(m[mode].alphabits+1);
		etc_expandv(palette[i][1], m[mode].colourbits+1, m[mode].colourbits+1, m[mode].colourbits+1); palette[i][0].v[3]|=palette[i][0].v[3]>>(m[mode].alphabits+1);
	}
	else
	{
		etc_expandv(palette[i][0], m[mode].colourbits, m[mode].colourbits, m[mode].colourbits); palette[i][0].v[3]|=palette[i][0].v[3]>>m[mode].alphabits;
		etc_expandv(palette[i][1], m[mode].colourbits, m[mode].colourbits, m[mode].colourbits);	palette[i][1].v[3]|=palette[i][0].v[3]>>m[mode].alphabits;
	}

	cb = m[mode].indexbits[idxsel];
	weight = wsz[cb];

	//build lerps table (could do it per pixel?)
	for (i = 0; i < m[mode].numsubsets; i++)
	{
		for (j = 0; j < (1<<cb); j++)
		{
			tab[i][j].v[0] = (palette[i][0].v[0]*(64-weight[j]) + palette[i][1].v[0]*weight[j] + 32)>>6;
			tab[i][j].v[1] = (palette[i][0].v[1]*(64-weight[j]) + palette[i][1].v[1]*weight[j] + 32)>>6;
			tab[i][j].v[2] = (palette[i][0].v[2]*(64-weight[j]) + palette[i][1].v[2]*weight[j] + 32)>>6;
			tab[i][j].v[3] = (palette[i][0].v[3]*(64-weight[j]) + palette[i][1].v[3]*weight[j] + 32)>>6;
		}

		//this stuff is annoying, but saves a bit or two
		if (!cb)
			anchor[i] = 16;	//don't allow the cb-1 to read negative bits!
		else if (i)
			anchor[i] = anchortable[(ss>2)?i:0][partition];
		else
			anchor[i] = 0;
	}

	//okay, tables are all set up, spew out the pixels
	for (i = 0; i < 16; )
	{
		int pidx = p[i];
		int idx;
		if (i == anchor[pidx])
			idx = ReadBits(in, &bit, cb-1);
		else
			idx = ReadBits(in, &bit, cb);
		out[i].u = tab[pidx][idx].u;
		i++;
		if (!(i & 3))
			out += w-4;
	}

	//mode has separate alpha indexes, spew those out too, clobbering any alpha from dodgy rgb blends
	if (m[mode].indexbits[idxsel^1])
	{	//FIXME: untested
		out -= w*4;
		cb = m[mode].indexbits[idxsel^1];
		weight = wsz[cb];
		for (i = 0; i < m[mode].numsubsets; i++)
		{
			for (j = 0; j < (1<<cb); j++)
				tab[i][j].v[3] = (palette[i][0].v[3]*(64-weight[j]) + palette[i][1].v[3]*weight[j] + 32)>>6;
		}

		for (i = 0; i < 16; )
		{
			int idx;
			if (i == 0)
				idx = ReadBits(in, &bit, cb-1);
			else
				idx = ReadBits(in, &bit, cb);
			out[i].v[3] = tab[p[i]][idx].v[3];
			i++;
			if (!(i & 3))
				out += w-4;
		}
	}

	//some modes allow swapping the alpha with an rgb channel (per block)
	if (rot)
	{	//FIXME: untested
		qbyte t;
		rot--; //0=disable, 1=red, 2=green, 3=blue
		out -= w*4;
		for (i = 0; i < 16; )
		{
			t = out[i].v[3];
			out[i].v[3] = out[i].v[rot];
			out[i].v[rot] = t;

			i++;
			if (!(i & 3))
				out += w-4;
		}
	}
}
#endif

static void Image_Decode_RGB8_Block(qbyte *fte_restrict in, pixel32_t *fte_restrict out, int w)
{
	Vector4Set(out->v, in[0], in[1], in[2], 0xff);
}
static void Image_Decode_L8A8_Block(qbyte *fte_restrict in, pixel32_t *fte_restrict out, int w)
{
	Vector4Set(out->v, in[0], in[0], in[0], in[1]);
}
static void Image_Decode_L8_Block(qbyte *fte_restrict in, pixel32_t *fte_restrict out, int w)
{
	Vector4Set(out->v, in[0], in[0], in[0], 0xff);
}

void Image_BlockSizeForEncoding(uploadfmt_t encoding, unsigned int *blockbytes, unsigned int *blockwidth, unsigned int *blockheight)
{
	unsigned int b = 0, w = 1, h = 1;
	switch(encoding)
	{
	case PTI_RGB565:
	case PTI_RGBA4444:
	case PTI_ARGB4444:
	case PTI_RGBA5551:
	case PTI_ARGB1555:
		b = 2;	//16bit formats
		break;
	case PTI_RGBA8:
	case PTI_RGBX8:
	case PTI_BGRA8:
	case PTI_BGRX8:
	case PTI_RGBA8_SRGB:
	case PTI_RGBX8_SRGB:
	case PTI_BGRA8_SRGB:
	case PTI_BGRX8_SRGB:
	case PTI_A2BGR10:
	case PTI_E5BGR9:
		b = 4;
		break;

	case PTI_R16:
	case PTI_R16F:
		b = 1*2;
		break;
	case PTI_R32F:
		b = 1*4;
		break;
	case PTI_RGBA16:
	case PTI_RGBA16F:
		b = 4*2;
		break;
	case PTI_RGBA32F:
		b = 4*4;
		break;
	case PTI_P8:
	case PTI_R8:
	case PTI_R8_SNORM:
		b = 1;
		break;
	case PTI_RG8:
	case PTI_RG8_SNORM:
		b = 2;
		break;

	case PTI_DEPTH16:
		b = 2;
		break;
	case PTI_DEPTH24:
		b = 3;
		break;
	case PTI_DEPTH32:
		b = 4;
		break;
	case PTI_DEPTH24_8:
		b = 4;
		break;

	case PTI_RGB8:
	case PTI_BGR8:
		b = 3;
		break;
	case PTI_L8:
	case PTI_L8_SRGB:
		b = 1;
		break;
	case PTI_L8A8:
	case PTI_L8A8_SRGB:
		b = 2;
		break;

	case PTI_BC1_RGB:
	case PTI_BC1_RGB_SRGB:
	case PTI_BC1_RGBA:
	case PTI_BC1_RGBA_SRGB:
	case PTI_BC4_R8:
	case PTI_BC4_R8_SNORM:
	case PTI_ETC1_RGB8:
	case PTI_ETC2_RGB8:
	case PTI_ETC2_RGB8_SRGB:
	case PTI_ETC2_RGB8A1:
	case PTI_ETC2_RGB8A1_SRGB:
	case PTI_EAC_R11:
	case PTI_EAC_R11_SNORM:
		w = h = 4;
		b = 8;
		break;
	case PTI_BC2_RGBA:
	case PTI_BC2_RGBA_SRGB:
	case PTI_BC3_RGBA:
	case PTI_BC3_RGBA_SRGB:
	case PTI_BC5_RG8:
	case PTI_BC5_RG8_SNORM:
	case PTI_BC6_RGB_UFLOAT:
	case PTI_BC6_RGB_SFLOAT:
	case PTI_BC7_RGBA:
	case PTI_BC7_RGBA_SRGB:
	case PTI_ETC2_RGB8A8:
	case PTI_ETC2_RGB8A8_SRGB:
	case PTI_EAC_RG11:
	case PTI_EAC_RG11_SNORM:
		w = h = 4;
		b = 16;
		break;

// ASTC is crazy with its format subtypes... note that all are potentially rgba, selected on a per-block basis
	case PTI_ASTC_4X4_SRGB:
	case PTI_ASTC_4X4:		w = 4; h = 4; b = 16; break;
	case PTI_ASTC_5X4_SRGB:
	case PTI_ASTC_5X4:		w = 5; h = 4; b = 16; break;
	case PTI_ASTC_5X5_SRGB:
	case PTI_ASTC_5X5:		w = 5; h = 5; b = 16; break;
	case PTI_ASTC_6X5_SRGB:
	case PTI_ASTC_6X5:		w = 6; h = 5; b = 16; break;
	case PTI_ASTC_6X6_SRGB:
	case PTI_ASTC_6X6:		w = 6; h = 6; b = 16; break;
	case PTI_ASTC_8X5_SRGB:
	case PTI_ASTC_8X5:		w = 8; h = 5; b = 16; break;
	case PTI_ASTC_8X6_SRGB:
	case PTI_ASTC_8X6:		w = 8; h = 6; b = 16; break;
	case PTI_ASTC_10X5_SRGB:
	case PTI_ASTC_10X5:		w = 10; h = 5; b = 16; break;
	case PTI_ASTC_10X6_SRGB:
	case PTI_ASTC_10X6:		w = 10; h = 6; b = 16; break;
	case PTI_ASTC_8X8_SRGB:
	case PTI_ASTC_8X8:		w = 8; h = 8; b = 16; break;
	case PTI_ASTC_10X8_SRGB:
	case PTI_ASTC_10X8:		w = 10; h = 8; b = 16; break;
	case PTI_ASTC_10X10_SRGB:
	case PTI_ASTC_10X10:	w = 10; h = 10; b = 16; break;
	case PTI_ASTC_12X10_SRGB:
	case PTI_ASTC_12X10:	w = 12; h = 10; b = 16; break;
	case PTI_ASTC_12X12_SRGB:
	case PTI_ASTC_12X12:	w = 12; h = 12; b = 16; break;

	case PTI_EMULATED:
#ifdef FTE_TARGET_WEB
	case PTI_WHOLEFILE: //UNKNOWN!
#endif
	case PTI_MAX:
		break;
	}

	*blockbytes = b;
	*blockwidth = w;
	*blockheight = h;
}

const char *Image_FormatName(uploadfmt_t fmt)
{
	switch(fmt)
	{
	case PTI_RGB565:			return "RGB565";
	case PTI_RGBA4444:			return "RGBA4444";
	case PTI_ARGB4444:			return "ARGB4444";
	case PTI_RGBA5551:			return "RGBA5551";
	case PTI_ARGB1555:			return "ARGB1555";
	case PTI_RGBA8:				return "RGBA8";
	case PTI_RGBX8:				return "RGBX8";
	case PTI_BGRA8:				return "BGRA8";
	case PTI_BGRX8:				return "BGRX8";
	case PTI_RGBA8_SRGB:		return "RGBA8_SRGB";
	case PTI_RGBX8_SRGB:		return "RGBX8_SRGB";
	case PTI_BGRA8_SRGB:		return "BGRA8_SRGB";
	case PTI_BGRX8_SRGB:		return "BGRX8_SRGB";
	case PTI_A2BGR10:			return "A2BGR10";
	case PTI_E5BGR9:			return "E5BGR9";
	case PTI_R16F:				return "R16F";
	case PTI_R32F:				return "R32F";
	case PTI_RGBA16F:			return "RGBA16F";
	case PTI_RGBA32F:			return "RGBA32F";
	case PTI_R16:				return "R16";
	case PTI_RGBA16:			return "RGBA16";
	case PTI_P8:				return "P8";
	case PTI_R8:				return "R8";
	case PTI_R8_SNORM:			return "R8_SNORM";
	case PTI_RG8:				return "RG8";
	case PTI_RG8_SNORM:			return "RG8_SNORM";
	case PTI_DEPTH16:			return "DEPTH16";
	case PTI_DEPTH24:			return "DEPTH24";
	case PTI_DEPTH32:			return "DEPTH32";
	case PTI_DEPTH24_8:			return "DEPTH24_8";
	case PTI_RGB8:				return "RGB8";
	case PTI_BGR8:				return "BGR8";
	case PTI_L8:				return "L8";
	case PTI_L8_SRGB:			return "L8_SRGB";
	case PTI_L8A8:				return "L8A8";
	case PTI_L8A8_SRGB:			return "L8A8_SRGB";
	case PTI_BC1_RGB:			return "BC1_RGB";
	case PTI_BC1_RGB_SRGB:		return "BC1_RGB_SRGB";
	case PTI_BC1_RGBA:			return "BC1_RGBA";
	case PTI_BC1_RGBA_SRGB:		return "BC1_RGBA_SRGB";
	case PTI_BC2_RGBA:			return "BC2_RGBA";
	case PTI_BC2_RGBA_SRGB:		return "BC2_RGBA_SRGB";
	case PTI_BC3_RGBA:			return "BC3_RGBA";
	case PTI_BC3_RGBA_SRGB:		return "BC3_RGBA_SRGB";
	case PTI_BC4_R8:			return "BC4_R8";
	case PTI_BC4_R8_SNORM:		return "BC4_R8_SNORM";
	case PTI_BC5_RG8:			return "BC5_RG8";
	case PTI_BC5_RG8_SNORM:		return "BC5_RG8_SNORM";
	case PTI_BC6_RGB_UFLOAT:	return "BC6_RGBF";
	case PTI_BC6_RGB_SFLOAT:	return "BC6_RGBF_SNORM";
	case PTI_BC7_RGBA:			return "BC7_RGBA";
	case PTI_BC7_RGBA_SRGB:		return "BC7_RGBA_SRGB";
	case PTI_ETC1_RGB8:			return "ETC1_RGB8";
	case PTI_ETC2_RGB8:			return "ETC2_RGB8";
	case PTI_ETC2_RGB8_SRGB:	return "ETC2_RGB8_SRGB";
	case PTI_ETC2_RGB8A1:		return "ETC2_RGB8A1";
	case PTI_ETC2_RGB8A1_SRGB:	return "ETC2_RGB8A1_SRGB";
	case PTI_EAC_R11:			return "EAC_R11";
	case PTI_EAC_R11_SNORM:		return "EAC_R11_SNORM";
	case PTI_ETC2_RGB8A8:		return "ETC2_RGB8A8";
	case PTI_ETC2_RGB8A8_SRGB:	return "ETC2_RGB8A8_SRGB";
	case PTI_EAC_RG11:			return "EAC_RG11";
	case PTI_EAC_RG11_SNORM:	return "EAC_RG11_SNORM";
	case PTI_ASTC_4X4_SRGB:		return "ASTC_4X4_SRGB";
	case PTI_ASTC_4X4:			return "ASTC_4X4";
	case PTI_ASTC_5X4_SRGB:		return "ASTC_5X4_SRGB";
	case PTI_ASTC_5X4:			return "ASTC_5X4";
	case PTI_ASTC_5X5_SRGB:		return "ASTC_5X5_SRGB";
	case PTI_ASTC_5X5:			return "ASTC_5X5";
	case PTI_ASTC_6X5_SRGB:		return "ASTC_6X5_SRGB";
	case PTI_ASTC_6X5:			return "ASTC_6X5";
	case PTI_ASTC_6X6_SRGB:		return "ASTC_6X6_SRGB";
	case PTI_ASTC_6X6:			return "ASTC_6X6";
	case PTI_ASTC_8X5_SRGB:		return "ASTC_8X5_SRGB";
	case PTI_ASTC_8X5:			return "ASTC_8X5";
	case PTI_ASTC_8X6_SRGB:		return "ASTC_8X6_SRGB";
	case PTI_ASTC_8X6:			return "ASTC_8X6";
	case PTI_ASTC_10X5_SRGB:	return "ASTC_10X5_SRGB";
	case PTI_ASTC_10X5:			return "ASTC_10X5";
	case PTI_ASTC_10X6_SRGB:	return "ASTC_10X6_SRGB";
	case PTI_ASTC_10X6:			return "ASTC_10X6";
	case PTI_ASTC_8X8_SRGB:		return "ASTC_8X8_SRGB";
	case PTI_ASTC_8X8:			return "ASTC_8X8";
	case PTI_ASTC_10X8_SRGB:	return "ASTC_10X8_SRGB";
	case PTI_ASTC_10X8:			return "ASTC_10X8";
	case PTI_ASTC_10X10_SRGB:	return "ASTC_10X10_SRGB";
	case PTI_ASTC_10X10:		return "ASTC_10X10";
	case PTI_ASTC_12X10_SRGB:	return "ASTC_12X10_SRGB";
	case PTI_ASTC_12X10:		return "ASTC_12X10";
	case PTI_ASTC_12X12_SRGB:	return "ASTC_12X12_SRGB";
	case PTI_ASTC_12X12:		return "ASTC_12X12";

#ifdef FTE_TARGET_WEB
	case PTI_WHOLEFILE:			return "Whole File";
#endif
	case PTI_EMULATED:
	case PTI_MAX:
		break;
	}
	return "Unknown";
}

static pixel32_t *Image_Block_Decode(qbyte *fte_restrict in, size_t insize, int w, int h, void(*decodeblock)(qbyte *fte_restrict in, pixel32_t *fte_restrict out, int w), uploadfmt_t encoding)
{
#define TMPBLOCKSIZE 16u
	pixel32_t *ret, *out;
	pixel32_t tmp[TMPBLOCKSIZE*TMPBLOCKSIZE];
	int x, y, i, j;
	int sizediff;

	unsigned int blockbytes, blockwidth, blockheight;
	Image_BlockSizeForEncoding(encoding, &blockbytes, &blockwidth, &blockheight);

	if (blockwidth > TMPBLOCKSIZE || blockheight > TMPBLOCKSIZE)
		Sys_Error("Image_Block_Decode only supports up to %u*%u blocks.\n", TMPBLOCKSIZE,TMPBLOCKSIZE);

	sizediff = insize - blockbytes*((w+blockwidth-1)/blockwidth)*((h+blockheight-1)/blockheight);
	if (sizediff)
	{
		Con_Printf("Image_Block_Decode: %s data size is %u, expected %u\n\n", Image_FormatName(encoding), (unsigned int)insize, (unsigned int)(insize-sizediff));
		if (sizediff < 0)
			return NULL;
	}

	ret = out = BZ_Malloc(w*h*sizeof(*out));

	for (y = 0; y < (h&~(blockheight-1)); y+=blockheight, out += w*(blockheight-1))
	{
		for (x = 0; x < (w&~(blockwidth-1)); x+=blockwidth, in+=blockbytes, out+=blockwidth)
			decodeblock(in, out, w);
		if (w%blockwidth)
		{
			decodeblock(in, tmp, TMPBLOCKSIZE);
			for (i = 0; x < w; x++, out++, i++)
			{
				for (j = 0; j < blockheight; j++)
					out[w*j] = tmp[i+TMPBLOCKSIZE*j];
			}
			in+=blockbytes;
		}
	}
	if (h%blockheight)
	{	//now walk along the bottom of the image
		h %= blockheight;
		for (x = 0; x < w; )
		{
			decodeblock(in, tmp, TMPBLOCKSIZE);
			i = 0;
			do
			{
				if (x == w)
					break;
				for (y = 0; y < h; y++)
					out[w*y] = tmp[i+TMPBLOCKSIZE*y];
				out++;
				i++;
			} while (++x % blockwidth);
			in+=blockbytes;
		}
	}
	return ret;
}
static pixel64_t *Image_Block_Decode64(qbyte *fte_restrict in, size_t insize, int w, int h, void(*decodeblock)(qbyte *fte_restrict in, pixel64_t *fte_restrict out, int w), uploadfmt_t encoding)
{
#define TMPBLOCKSIZE 16u
	pixel64_t *ret, *out;
	pixel64_t tmp[TMPBLOCKSIZE*TMPBLOCKSIZE];
	int x, y, i, j;
	int sizediff;

	unsigned int blockbytes, blockwidth, blockheight;
	Image_BlockSizeForEncoding(encoding, &blockbytes, &blockwidth, &blockheight);

	if (blockwidth > TMPBLOCKSIZE || blockheight > TMPBLOCKSIZE)
		Sys_Error("Image_Block_Decode only supports up to %u*%u blocks.\n", TMPBLOCKSIZE,TMPBLOCKSIZE);

	sizediff = insize - blockbytes*((w+blockwidth-1)/blockwidth)*((h+blockheight-1)/blockheight);
	if (sizediff)
	{
		Con_Printf("Image_Block_Decode: %s data size is %u, expected %u\n\n", Image_FormatName(encoding), (unsigned int)insize, (unsigned int)(insize-sizediff));
		if (sizediff < 0)
			return NULL;
	}

	ret = out = BZ_Malloc(w*h*sizeof(*out));

	for (y = 0; y < (h&~(blockheight-1)); y+=blockheight, out += w*(blockheight-1))
	{
		for (x = 0; x < (w&~(blockwidth-1)); x+=blockwidth, in+=blockbytes, out+=blockwidth)
			decodeblock(in, out, w);
		if (w%blockwidth)
		{
			decodeblock(in, tmp, TMPBLOCKSIZE);
			for (i = 0; x < w; x++, out++, i++)
			{
				for (j = 0; j < blockheight; j++)
					out[w*j] = tmp[i+TMPBLOCKSIZE*j];
			}
			in+=blockbytes;
		}
	}
	if (h%blockheight)
	{	//now walk along the bottom of the image
		h %= blockheight;
		for (x = 0; x < w; )
		{
			decodeblock(in, tmp, TMPBLOCKSIZE);
			i = 0;
			do
			{
				if (x == w)
					break;
				for (y = 0; y < h; y++)
					out[w*y] = tmp[i+TMPBLOCKSIZE*y];
				out++;
				i++;
			} while (++x % blockwidth);
			in+=blockbytes;
		}
	}
	return ret;
}

static void Image_DecompressFormat(struct pendingtextureinfo *mips)
{
	//various compressed formats might not be supported by various gpus/apis.
	//sometimes the gpu might only partially support the format (eg: d3d requires mip 0 be a multiple of the block size)
	//and sometimes we want the actual rgb data (eg: so that we can palettize it)
	//so this is still useful even if every driver ever created supported the format.
	//as a general rule, decompressing is fairly straight forward, but not free. yay threads.

	//iiuc any basic s3tc patents have now expired, so it is legally safe to decode (though fancy compression logic may still have restrictions, but we don't compress).
	static float throttle;
	void (*decodefunc)(qbyte *fte_restrict, pixel32_t *fte_restrict, int) = NULL;
	void (*decodefunc64)(qbyte *fte_restrict, pixel64_t *fte_restrict, int) = NULL;
	int rcoding = mips->encoding;
	int mip;
	switch(mips->encoding)
	{
	default:
		break;
	case PTI_RGB8:
		decodefunc = Image_Decode_RGB8_Block;
		rcoding = PTI_RGBX8;
		break;
	case PTI_L8A8:
		decodefunc = Image_Decode_L8A8_Block;
		rcoding = PTI_RGBA8;
		break;
	case PTI_L8:
		decodefunc = Image_Decode_L8_Block;
		rcoding = PTI_RGBA8;
		break;
#ifdef DECOMPRESS_ETC2
	case PTI_ETC1_RGB8:
	case PTI_ETC2_RGB8: //backwards compatible, so we just treat them the same
	case PTI_ETC2_RGB8_SRGB:
		decodefunc = Image_Decode_ETC2_RGB8_Block;
		rcoding = (mips->encoding==PTI_ETC2_RGB8_SRGB)?PTI_RGBX8_SRGB:PTI_RGBX8;
		break;
	case PTI_ETC2_RGB8A1:	//weird hack mode
	case PTI_ETC2_RGB8A1_SRGB:
		decodefunc = Image_Decode_ETC2_RGB8A1_Block;
		rcoding = (mips->encoding==PTI_ETC2_RGB8A1_SRGB)?PTI_RGBA8_SRGB:PTI_RGBA8;
		break;
	case PTI_ETC2_RGB8A8:
	case PTI_ETC2_RGB8A8_SRGB:
		decodefunc = Image_Decode_ETC2_RGB8A8_Block;
		rcoding = (mips->encoding==PTI_ETC2_RGB8A8_SRGB)?PTI_RGBA8_SRGB:PTI_RGBA8;
		break;
	case PTI_EAC_R11:
		decodefunc = Image_Decode_EAC_R11U_Block;
		rcoding = PTI_RGBX8;
		break;
/*	case PTI_EAC_R11_SNORM:
		decodefunc = Image_Decode_EAC_R11S_Block;
		rcoding = PTI_RGBX8;
		break;*/
	case PTI_EAC_RG11:
		decodefunc = Image_Decode_EAC_RG11U_Block;
		rcoding = PTI_RGBX8;
		break;
/*	case PTI_EAC_RG11_SNORM:
		decodefunc = Image_Decode_EAC_RG11S_Block;
		rcoding = PTI_RGBX8;
		break;*/
#else
	case PTI_ETC1_RGB8:
	case PTI_ETC2_RGB8:
	case PTI_ETC2_RGB8_SRGB:
	case PTI_ETC2_RGB8A1:
	case PTI_ETC2_RGB8A1_SRGB:
	case PTI_ETC2_RGB8A8:
	case PTI_ETC2_RGB8A8_SRGB:
	case PTI_EAC_R11:
	case PTI_EAC_R11_SNORM:
	case PTI_EAC_RG11:
	case PTI_EAC_RG11_SNORM:
		Con_ThrottlePrintf(&throttle, 0, "ETC1/ETC2/EAC decompression is not supported in this build\n");
		break;
#endif

	case PTI_BC1_RGB:
	case PTI_BC1_RGB_SRGB:
#ifdef DECOMPRESS_S3TC
		decodefunc = Image_Decode_BC1_Block;
		rcoding = (mips->encoding==PTI_BC1_RGB_SRGB)?PTI_RGBX8_SRGB:PTI_RGBX8;
#else
		Con_ThrottlePrintf(&throttle, 0, "BC1 decompression is not supported in this build\n");
#endif
		break;
	case PTI_BC1_RGBA:
	case PTI_BC1_RGBA_SRGB:
#ifdef DECOMPRESS_S3TC
		decodefunc = Image_Decode_BC1A_Block;
		rcoding = (mips->encoding==PTI_BC1_RGBA_SRGB)?PTI_RGBA8_SRGB:PTI_RGBA8;
#else
		Con_ThrottlePrintf(&throttle, 0, "BC1A decompression is not supported in this build\n");
#endif
		break;
	case PTI_BC2_RGBA:
	case PTI_BC2_RGBA_SRGB:
#ifdef DECOMPRESS_S3TC
		decodefunc = Image_Decode_BC2_Block;
		rcoding = (mips->encoding==PTI_BC2_RGBA_SRGB)?PTI_RGBA8_SRGB:PTI_RGBA8;
#else
		Con_ThrottlePrintf(&throttle, 0, "BC2 decompression is not supported in this build\n");
#endif
		break;
	case PTI_BC3_RGBA:
	case PTI_BC3_RGBA_SRGB:
#if defined(DECOMPRESS_RGTC) && defined(DECOMPRESS_S3TC)
		decodefunc = Image_Decode_BC3_Block;
		rcoding = (mips->encoding==PTI_BC3_RGBA_SRGB)?PTI_RGBA8_SRGB:PTI_RGBA8;
#else
		Con_ThrottlePrintf(&throttle, 0, "BC3 decompression is not supported in this build\n");
#endif
		break;
#ifdef DECOMPRESS_RGTC
	case PTI_BC4_R8_SNORM:
		decodefunc = Image_Decode_BC4S_Block;
		rcoding = PTI_RGBX8;
		break;
	case PTI_BC4_R8:
		decodefunc = Image_Decode_BC4U_Block;
		rcoding = PTI_RGBX8;
		break;
	case PTI_BC5_RG8_SNORM:
		decodefunc = Image_Decode_BC5S_Block;
		rcoding = PTI_RGBX8;
		break;
	case PTI_BC5_RG8:
		decodefunc = Image_Decode_BC5U_Block;
		rcoding = PTI_RGBX8;
		break;
#else
	case PTI_BC4_R8_SNORM:
	case PTI_BC4_R8:
	case PTI_BC5_RG8_SNORM:
	case PTI_BC5_RG8:
		Con_ThrottlePrintf(&throttle, 0, "BC4/BC5 decompression is not supported in this build\n");
		break;
#endif
	case PTI_BC6_RGB_UFLOAT:
#ifdef DECOMPRESS_BPTC
		decodefunc64 = Image_Decode_BC6U_Block;
		rcoding = PTI_RGBA16F;
#else
		Con_ThrottlePrintf(&throttle, 0, "BC6_UFLOAT decompression is not supported\n");
#endif
		break;
	case PTI_BC6_RGB_SFLOAT:
#ifdef DECOMPRESS_BPTC
		decodefunc64 = Image_Decode_BC6S_Block;
		rcoding = PTI_RGBA16F;
#else
		Con_ThrottlePrintf(&throttle, 0, "BC6_SFLOAT decompression is not supported\n");
#endif
		break;
	case PTI_BC7_RGBA:
	case PTI_BC7_RGBA_SRGB:
#ifdef DECOMPRESS_BPTC
		decodefunc = Image_Decode_BC7_Block;
		rcoding = (mips->encoding==PTI_BC7_RGBA_SRGB)?PTI_RGBA8_SRGB:PTI_RGBA8;
#else
		Con_ThrottlePrintf(&throttle, 0, "BC7 decompression is not supported\n");
#endif
		break;
	case PTI_INVALID:
		Con_ThrottlePrintf(&throttle, 0, "Attempting to decompress invalid format\n");
		break;
	}
	if (decodefunc || decodefunc64)
	{
		for (mip = 0; mip < mips->mipcount; mip++)
		{
			size_t sz;
			void *out;
			if (decodefunc64)
			{
				sz = sizeof(pixel64_t);
				out = Image_Block_Decode64(mips->mip[mip].data, mips->mip[mip].datasize, mips->mip[mip].width, mips->mip[mip].height, decodefunc64, mips->encoding);
			}
			else
			{
				sz = sizeof(pixel32_t);
				out = Image_Block_Decode(mips->mip[mip].data, mips->mip[mip].datasize, mips->mip[mip].width, mips->mip[mip].height, decodefunc, mips->encoding);
			}
			if (mips->mip[mip].needfree)
				BZ_Free(mips->mip[mip].data);
			mips->mip[mip].data = out;
			mips->mip[mip].needfree = true;
			mips->mip[mip].datasize = mips->mip[mip].width*mips->mip[mip].height*sz;
		}
		if (mips->extrafree)
			BZ_Free(mips->extrafree);	//might as well free this now, as nothing is poking it any more.
		mips->extrafree = NULL;
		mips->encoding = rcoding;
	}
}

static void Image_ChangeFormat(struct pendingtextureinfo *mips, unsigned int flags, uploadfmt_t origfmt)
{
	int mip;

	if (mips->type != PTI_2D)
		return;	//blurgh

	if (flags & IF_PALETTIZE)
	{
		Image_DecompressFormat(mips);	//force-decompress it, so that we can palettise it.
		if (mips->encoding == PTI_RGBX8 || mips->encoding == PTI_RGBA8)
		{
			mips->encoding = PTI_P8;
			for (mip = 0; mip < mips->mipcount; mip++)
			{
				unsigned int i;
				unsigned char *out;
				unsigned char *in;
				void *needfree = NULL;

				in = mips->mip[mip].data;
				if (mips->mip[mip].needfree)
					out = in;
				else
				{
					needfree = in;
					out = BZ_Malloc(mips->mip[mip].width*mips->mip[mip].height*sizeof(*out));
					mips->mip[mip].data = out;
				}
				mips->mip[mip].datasize = mips->mip[mip].width*mips->mip[mip].height;
				mips->mip[mip].needfree = true;

				for (i = 0; i < mips->mip[mip].width*mips->mip[mip].height; i++, in+=4)
					out[i] = GetPaletteIndexNoFB(in[0], in[1], in[2]);

				if (needfree)
					BZ_Free(needfree);
			}
		}
	}

	//if that format isn't supported/desired, try converting it.
	if (sh_config.texfmt[mips->encoding])
	{
		if (!sh_config.texture_allow_block_padding && mips->mipcount)
		{	//direct3d is annoying, and will reject any block-compressed format with a base mip size that is not a multiple of the block size.
			//its fine with weirdly sized mips though. I have no idea why there's this restriction, but whatever.
			//we need to de
			int blockbytes, blockwidth, blockheight;
			Image_BlockSizeForEncoding(mips->encoding, &blockbytes, &blockwidth, &blockheight);
			if (!(mips->mip[0].width % blockwidth) && !(mips->mip[0].height % blockheight))
				return;
			//else encoding isn't supported for this size. fall through.
		}
		else
			return;
	}

	//when the format can't be used, decompress it if its one of those awkward compressed formats.
	Image_DecompressFormat(mips);
	if (sh_config.texfmt[mips->encoding])
		return;	//okay, that got it.

	//Fixme: PTI_E5BGR9 -> PTI_RGBA32F
	//Fixme: PTI_RGBA16F -> PTI_RGBA32F
	//FIXME: PTI_RGBA32F -> PTI_RGBA8

	if (mips->encoding == PTI_RGBA16)
	{
		Image_4X16to8888(mips);
		mips->encoding = PTI_RGBA8;
		if (sh_config.texfmt[mips->encoding])
			return;	//okay, that got it.
	}

	if ((mips->encoding == PTI_RGBX8 && sh_config.texfmt[PTI_BGRX8]) ||
		(mips->encoding == PTI_BGRX8 && sh_config.texfmt[PTI_BGRX8]))
	{
		Image_32To24(mips);
		mips->encoding = (mips->encoding == PTI_RGBX8)?PTI_RGB8:PTI_BGR8;
	}
	else if ((mips->encoding == PTI_RGBX8 && sh_config.texfmt[PTI_BGRX8]) ||
		(mips->encoding == PTI_BGRX8 && sh_config.texfmt[PTI_RGBX8]) ||
		(mips->encoding == PTI_RGBA8 && sh_config.texfmt[PTI_BGRA8]) ||
		(mips->encoding == PTI_BGRA8 && sh_config.texfmt[PTI_RGBA8]))
	{
		for (mip = 0; mip < mips->mipcount; mip++)
			Image_8_BGR_RGB_Swap(mips->mip[mip].data, mips->mip[mip].width, mips->mip[mip].height);
		if (mips->encoding == PTI_RGBA8)
			mips->encoding = PTI_BGRA8;
		else if (mips->encoding == PTI_BGRA8)
			mips->encoding = PTI_RGBA8;
		else if (mips->encoding == PTI_RGBX8)
			mips->encoding = PTI_BGRX8;
		else// if (mips->encoding == PTI_BGRX8)
			mips->encoding = PTI_RGBX8;
	}
	//should we just use 5551 always?
	else if (mips->encoding == PTI_RGBX8 || mips->encoding == PTI_BGRX8)
	{
		/*if (0)
		{	//prevent discolouration.
			if (sh_config.texfmt[PTI_RGBA5551])
			{
				for (mip = 0; mip < mips->mipcount; mip++)
					Image_8888to5551(mips, mip, mips->encoding == PTI_BGRX8);
				mips->encoding = PTI_RGBA5551;
			}
			else
			{
				for (mip = 0; mip < mips->mipcount; mip++)
					Image_8888to1555(mips, mip, mips->encoding == PTI_BGRX8);
				mips->encoding = PTI_ARGB1555;
			}
		}
		else*/
		if (sh_config.texfmt[PTI_RGB565])
		{
			Image_8888to565(mips, mips->encoding == PTI_BGRX8);
			mips->encoding = PTI_RGB565;
		}
	}
	else if (mips->encoding == PTI_RGBA8 || mips->encoding == PTI_BGRA8)
	{
		if (origfmt == TF_TRANS8 || origfmt == TF_TRANS8_FULLBRIGHT || origfmt == TF_H2_TRANS8_0 || !(sh_config.texfmt[PTI_RGBA4444] || sh_config.texfmt[PTI_ARGB4444]))
		{	//1-bit alpha is okay for these textures.
			if (sh_config.texfmt[PTI_RGBA5551])
			{
				Image_8888to5551(mips, mips->encoding == PTI_BGRA8);
				mips->encoding = PTI_RGBA5551;
			}
			else
			{
				Image_8888to1555(mips, mips->encoding == PTI_BGRA8);
				mips->encoding = PTI_ARGB1555;
			}
		}
		else
		{
			if (sh_config.texfmt[PTI_RGBA4444])
			{
				Image_8888to4444(mips, mips->encoding == PTI_BGRA8);
				mips->encoding = PTI_RGBA4444;
			}
			else
			{
				Image_8888toARGB4444(mips, mips->encoding == PTI_BGRA8);
				mips->encoding = PTI_ARGB4444;
			}
		}
	}
}

//resamples and depalettes as required
//ALWAYS frees rawdata, even on failure (but never mips).
static qboolean Image_GenMip0(struct pendingtextureinfo *mips, unsigned int flags, void *rawdata, void *palettedata, int imgwidth, int imgheight, uploadfmt_t fmt, qboolean freedata)
{
	unsigned int *rgbadata = rawdata;
	int i;
	qboolean valid;
	unsigned int bb, bw, bh;

	mips->mip[0].width = imgwidth;
	mips->mip[0].height = imgheight;
	mips->mip[0].depth = 1;
	mips->mipcount = 1;

	switch(fmt)
	{
	default:
		mips->encoding = fmt;
		break;

	case TF_INVALID:
		Con_Printf("R_LoadRawTexture: bad format\n");
		if (freedata)
			BZ_Free(rawdata);
		return false;

	case TF_MIP4_P8:
		//8bit indexed data.
		Image_RoundDimensions(&mips->mip[0].width, &mips->mip[0].height, flags);
		flags |= IF_NOPICMIP;
		if (/*!r_dodgymiptex.ival &&*/ mips->mip[0].width == imgwidth && mips->mip[0].height == imgheight)
		{
			unsigned int pixels =
				(imgwidth>>0) * (imgheight>>0) + 
				(imgwidth>>1) * (imgheight>>1) +
				(imgwidth>>2) * (imgheight>>2) +
				(imgwidth>>3) * (imgheight>>3);

			mips->encoding = PTI_P8;
			rgbadata = BZ_Malloc(pixels);
			memcpy(rgbadata, rawdata, pixels);

			for (i = 0; i < 4; i++)
			{
				mips->mip[i].width = imgwidth>>i;
				mips->mip[i].height = imgheight>>i;
				mips->mip[i].depth = 1;
				mips->mip[i].datasize = mips->mip[i].width * mips->mip[i].height;
				mips->mip[i].needfree = false;
			}
			mips->mipcount = i;
			mips->mip[0].data = rgbadata;
			mips->mip[1].data = (qbyte*)mips->mip[0].data + mips->mip[0].datasize;
			mips->mip[2].data = (qbyte*)mips->mip[1].data + mips->mip[1].datasize;
			mips->mip[3].data = (qbyte*)mips->mip[2].data + mips->mip[2].datasize;

			mips->extrafree = rgbadata;
			if (freedata)
				BZ_Free(rawdata);
			return true;
		}
		//fall through
	case PTI_LLLX8:
		if (sh_config.texfmt[((vid.flags & VID_SRGBAWARE) /*&& (flags & IF_SRGB)*/ && !(flags & IF_NOSRGB))?PTI_L8_SRGB:PTI_L8])
		{	//if we can compact it, then do so!
			mips->encoding = PTI_L8;
			//can just do this in-place.
			for (i = 0; i < imgwidth * imgheight; i++)
				((qbyte*)rgbadata)[i] = ((qbyte*)rgbadata)[i*4];
		}
		//otherwise treat it as whatever the gpu prefers
		else if (sh_config.texfmt[PTI_BGRX8] || sh_config.texfmt[PTI_BGRA8])
			mips->encoding = PTI_BGRX8;
		else
			mips->encoding = PTI_RGBX8;
		break;
	case PTI_LLLA8:
		//take special care here, because L8A8_SRGB doesn't exist in core gl, nor can it be easily faked.
		if (sh_config.texfmt[((vid.flags & VID_SRGBAWARE) /*&& (flags & IF_SRGB)*/ && !(flags & IF_NOSRGB))?PTI_L8A8_SRGB:PTI_L8A8])
		{	//if we can compact it, then do so!
			mips->encoding = PTI_L8A8;
			//can just do this in-place.
			for (i = 0; i < imgwidth * imgheight; i++)
			{
				((qbyte*)rgbadata)[i*2+0] = ((qbyte*)rgbadata)[i*4+0];
				((qbyte*)rgbadata)[i*2+1] = ((qbyte*)rgbadata)[i*4+3];
			}
		}
		else if (sh_config.texfmt[PTI_BGRA8])
			mips->encoding = PTI_BGRA8;
		else
			mips->encoding = PTI_RGBA8;
		break;
	case TF_MIP4_SOLID8:
		//8bit opaque data
		Image_RoundDimensions(&mips->mip[0].width, &mips->mip[0].height, flags);
		flags |= IF_NOPICMIP;
		if (!r_dodgymiptex.ival && mips->mip[0].width == imgwidth && mips->mip[0].height == imgheight)
		{	//special hack required to preserve the hand-drawn lower mips.
			unsigned int pixels =
				(imgwidth>>0) * (imgheight>>0) + 
				(imgwidth>>1) * (imgheight>>1) +
				(imgwidth>>2) * (imgheight>>2) +
				(imgwidth>>3) * (imgheight>>3);

			mips->encoding = PTI_RGBX8;
			rgbadata = BZ_Malloc(pixels*4);
			for (i = 0; i < pixels; i++)
				rgbadata[i] = d_8to24rgbtable[((qbyte*)rawdata)[i]];

			for (i = 0; i < 4; i++)
			{
				mips->mip[i].width = imgwidth>>i;
				mips->mip[i].height = imgheight>>i;
				mips->mip[i].depth = 1;
				mips->mip[i].datasize = mips->mip[i].width * mips->mip[i].height * 4;
				mips->mip[i].needfree = false;
			}
			mips->mipcount = i;
			mips->mip[0].data = rgbadata;
			mips->mip[1].data = (qbyte*)mips->mip[0].data + mips->mip[0].datasize;
			mips->mip[2].data = (qbyte*)mips->mip[1].data + mips->mip[1].datasize;
			mips->mip[3].data = (qbyte*)mips->mip[2].data + mips->mip[2].datasize;

			mips->extrafree = rgbadata;
			if (freedata)
				BZ_Free(rawdata);
			return true;
		}
		//fall through
	case TF_SOLID8:
		rgbadata = BZ_Malloc(imgwidth * imgheight*4);
		if (sh_config.texfmt[PTI_BGRX8])
		{	//bgra8 is typically faster when supported.
			mips->encoding = PTI_BGRX8;
			for (i = 0; i < imgwidth * imgheight; i++)
				rgbadata[i] = d_8to24bgrtable[((qbyte*)rawdata)[i]];
		}
		else
		{
			mips->encoding = PTI_RGBX8;
			for (i = 0; i < imgwidth * imgheight; i++)
				rgbadata[i] = d_8to24rgbtable[((qbyte*)rawdata)[i]];
		}
		if (freedata)
			BZ_Free(rawdata);
		freedata = true;
		break;
	case TF_TRANS8:
		{
			mips->encoding = PTI_RGBX8;
			rgbadata = BZ_Malloc(imgwidth * imgheight*4);
			for (i = 0; i < imgwidth * imgheight; i++)
			{
				if (((qbyte*)rawdata)[i] == 0xff)
				{//fixme: blend non-0xff neighbours. no, just use premultiplied alpha instead, where it matters.
					rgbadata[i] = 0;
					mips->encoding = PTI_RGBA8;
				}
				else
					rgbadata[i] = d_8to24rgbtable[((qbyte*)rawdata)[i]];
			}
			if (freedata)
				BZ_Free(rawdata);
			freedata = true;
		}
		break;
	case TF_H2_TRANS8_0:
		{
			mips->encoding = PTI_RGBX8;
			rgbadata = BZ_Malloc(imgwidth * imgheight*4);
			for (i = 0; i < imgwidth * imgheight; i++)
			{
				if (((qbyte*)rawdata)[i] == 0xff || ((qbyte*)rawdata)[i] == 0)
				{//fixme: blend non-0xff neighbours. no, just use premultiplied alpha instead, where it matters.
					rgbadata[i] = 0;
					mips->encoding = PTI_RGBA8;
				}
				else
					rgbadata[i] = d_8to24rgbtable[((qbyte*)rawdata)[i]];
			}
			if (freedata)
				BZ_Free(rawdata);
			freedata = true;
		}
		break;
	case TF_TRANS8_FULLBRIGHT:
		mips->encoding = PTI_RGBA8;
		rgbadata = BZ_Malloc(imgwidth * imgheight*4);
		for (i = 0, valid = false; i < imgwidth * imgheight; i++)
		{
			if (((qbyte*)rawdata)[i] == 255 || ((qbyte*)rawdata)[i] < 256-vid.fullbright)
				rgbadata[i] = 0;
			else
			{
				rgbadata[i] = d_8to24rgbtable[((qbyte*)rawdata)[i]];
				valid = true;
			}
		}
		if (freedata)
			BZ_Free(rawdata);
		freedata = true;
		if (!valid)
		{
			BZ_Free(rgbadata);
			return false;
		}
		break;

	case TF_HEIGHT8PAL:
		mips->encoding = PTI_RGBA8;
		rgbadata = BZ_Malloc(imgwidth * imgheight*5);
		{
			qbyte *heights = (qbyte*)(rgbadata + (imgwidth*imgheight));
			for (i = 0; i < imgwidth * imgheight; i++)
			{
				unsigned int rgb = d_8to24rgbtable[((qbyte*)rawdata)[i]];
				heights[i] = (((rgb>>16)&0xff) + ((rgb>>8)&0xff) + ((rgb>>0)&0xff))/3;
			}
			Image_GenerateNormalMap(heights, rgbadata, imgwidth, imgheight, r_shadow_bumpscale_basetexture.value?r_shadow_bumpscale_basetexture.value:4, r_shadow_heightscale_basetexture.value);
		}
		if (freedata)
			BZ_Free(rawdata);
		freedata = true;
		break;
	case TF_HEIGHT8:
		mips->encoding = PTI_RGBA8;
		rgbadata = BZ_Malloc(imgwidth * imgheight*4);
		Image_GenerateNormalMap(rawdata, rgbadata, imgwidth, imgheight, r_shadow_bumpscale_bumpmap.value, r_shadow_heightscale_bumpmap.value);
		if (freedata)
			BZ_Free(rawdata);
		freedata = true;
		break;

	case TF_BGR24_FLIP:
		mips->encoding = PTI_RGBX8;
		rgbadata = BZ_Malloc(imgwidth * imgheight*4);
		for (i = 0; i < imgheight; i++)
		{
			int x;
			qbyte *in = (qbyte*)rawdata + (imgheight-i-1) * imgwidth * 3;
			qbyte *out = (qbyte*)rgbadata + i * imgwidth * 4;
			for (x = 0; x < imgwidth; x++, in+=3, out+=4)
			{
				out[0] = in[2];
				out[1] = in[1];
				out[2] = in[0];
				out[3] = 0xff;
			}
		}
		if (freedata)
			BZ_Free(rawdata);
		freedata = true;
		break;

	case TF_MIP4_8PAL24_T255:
	case TF_MIP4_8PAL24:
		//8bit opaque data
		{
			unsigned int pixels =
					(imgwidth>>0) * (imgheight>>0) + 
					(imgwidth>>1) * (imgheight>>1) +
					(imgwidth>>2) * (imgheight>>2) +
					(imgwidth>>3) * (imgheight>>3);
			palettedata = (qbyte*)rawdata + pixels;
			Image_RoundDimensions(&mips->mip[0].width, &mips->mip[0].height, flags);
			flags |= IF_NOPICMIP;
			if (!r_dodgymiptex.ival && mips->mip[0].width == imgwidth && mips->mip[0].height == imgheight)
			{
				unsigned int pixels =
					(imgwidth>>0) * (imgheight>>0) + 
					(imgwidth>>1) * (imgheight>>1) +
					(imgwidth>>2) * (imgheight>>2) +
					(imgwidth>>3) * (imgheight>>3);

				rgbadata = BZ_Malloc(pixels*4);
				if (fmt == TF_MIP4_8PAL24_T255)
				{
					mips->encoding = PTI_RGBA8;
					for (i = 0; i < pixels; i++)
					{
						qbyte idx = ((qbyte*)rawdata)[i];
						if (idx == 255)
							rgbadata[i] = 0;
						else
						{
							qbyte *p = ((qbyte*)palettedata) + idx*3;
							rgbadata[i] = 0xff000000 | (p[0]<<0) | (p[1]<<8) | (p[2]<<16);	//FIXME: endian
						}
					}
				}
				else
				{
					mips->encoding = PTI_RGBX8;
					for (i = 0; i < pixels; i++)
					{
						qbyte *p = ((qbyte*)palettedata) + ((qbyte*)rawdata)[i]*3;
						//FIXME: endian
						rgbadata[i] = 0xff000000 | (p[0]<<0) | (p[1]<<8) | (p[2]<<16);
					}
				}

				for (i = 0; i < 4; i++)
				{
					mips->mip[i].width = imgwidth>>i;
					mips->mip[i].height = imgheight>>i;
					mips->mip[i].depth = 1;
					mips->mip[i].datasize = mips->mip[i].width * mips->mip[i].height * 4;
					mips->mip[i].needfree = false;
				}
				mips->mipcount = i;
				mips->mip[0].data = rgbadata;
				mips->mip[1].data = (qbyte*)mips->mip[0].data + mips->mip[0].datasize;
				mips->mip[2].data = (qbyte*)mips->mip[1].data + mips->mip[1].datasize;
				mips->mip[3].data = (qbyte*)mips->mip[2].data + mips->mip[2].datasize;

				mips->extrafree = rgbadata;
				if (freedata)
					BZ_Free(rawdata);
				return true;
			}
		}
		//fall through
	case TF_8PAL24:
		if (!palettedata)
		{
			Con_Printf("TF_8PAL24: no palette");
			if (freedata)
				BZ_Free(rawdata);
			return false;
		}
		rgbadata = BZ_Malloc(imgwidth * imgheight*4);
		if (fmt == TF_MIP4_8PAL24_T255)
		{
			mips->encoding = PTI_RGBA8;
			for (i = 0; i < imgwidth * imgheight; i++)
			{
				qbyte idx = ((qbyte*)rawdata)[i];
				if (idx == 255)
					rgbadata[i] = 0;
				else
				{
					qbyte *p = ((qbyte*)palettedata) + idx*3;
					rgbadata[i] = 0xff000000 | (p[0]<<0) | (p[1]<<8) | (p[2]<<16);	//FIXME: endian
				}
			}
		}
		else
		{
			mips->encoding = PTI_RGBX8;
			for (i = 0; i < imgwidth * imgheight; i++)
			{
				qbyte *p = ((qbyte*)palettedata) + ((qbyte*)rawdata)[i]*3;
				//FIXME: endian
				rgbadata[i] = 0xff000000 | (p[0]<<0) | (p[1]<<8) | (p[2]<<16);
			}
		}
		if (freedata)
			BZ_Free(rawdata);
		freedata = true;
		break;
	case TF_8PAL32:
		if (!palettedata)
		{
			Con_Printf("TF_8PAL32: no palette");
			if (freedata)
				BZ_Free(rawdata);
			return false;
		}
		mips->encoding = PTI_RGBA8;
		rgbadata = BZ_Malloc(imgwidth * imgheight*4);
		for (i = 0; i < imgwidth * imgheight; i++)
			rgbadata[i] = ((unsigned int*)palettedata)[((qbyte*)rawdata)[i]];
		if (freedata)
			BZ_Free(rawdata);
		freedata = true;
		break;

#ifdef HEXEN2
	case TF_H2_T7G1: /*8bit data, odd indexes give greyscale transparence*/
		mips->encoding = PTI_RGBA8;
		rgbadata = BZ_Malloc(imgwidth * imgheight*4);
		for (i = 0; i < imgwidth * imgheight; i++)
		{
			qbyte p = ((qbyte*)rawdata)[i];
			rgbadata[i] = d_8to24rgbtable[p] & 0x00ffffff;
			if (p == 0)
				;
			else if (p&1)
				rgbadata[i] |= 0x80000000;
			else
				rgbadata[i] |= 0xff000000;
		}
		if (freedata)
			BZ_Free(rawdata);
		freedata = true;
		break;
	case TF_H2_T4A4:     /*8bit data, weird packing*/
		mips->encoding = PTI_RGBA8;
		rgbadata = BZ_Malloc(imgwidth * imgheight*4);
		for (i = 0; i < imgwidth * imgheight; i++)
		{
			static const int ColorIndex[16] = {0, 31, 47, 63, 79, 95, 111, 127, 143, 159, 175, 191, 199, 207, 223, 231};
			static const unsigned ColorPercent[16] = {25, 51, 76, 102, 114, 127, 140, 153, 165, 178, 191, 204, 216, 229, 237, 247};
			qbyte p = ((qbyte*)rawdata)[i];
			rgbadata[i] = d_8to24rgbtable[ColorIndex[p>>4]] & 0x00ffffff;
			rgbadata[i] |= ( int )ColorPercent[p&15] << 24;
		}
		if (freedata)
			BZ_Free(rawdata);
		freedata = true;
		break;
#endif
	}

	if (flags & IF_NOALPHA)
	{
		switch(mips->encoding)
		{
		case PTI_RGBA8:
			mips->encoding = PTI_RGBX8;
			break;
		case PTI_BGRA8:
			mips->encoding = PTI_BGRX8;
			break;
		case PTI_RGBA8_SRGB:
			mips->encoding = PTI_RGBX8_SRGB;
			break;
		case PTI_BGRA8_SRGB:
			mips->encoding = PTI_BGRX8_SRGB;
			break;
		case PTI_RGBA16:
		case PTI_RGBA16F:
		case PTI_RGBA32F:
		case PTI_ARGB4444:
		case PTI_ARGB1555:
		case PTI_RGBA4444:
		case PTI_RGBA5551:
		case PTI_A2BGR10:
		case PTI_L8A8: //could strip.
		case PTI_L8A8_SRGB: //could strip.
			break;	//erk
		case PTI_BC1_RGBA:
			mips->encoding = PTI_BC1_RGB;
			break;
		case PTI_BC1_RGBA_SRGB:
			mips->encoding = PTI_BC1_RGB_SRGB;
			break;
		case PTI_BC2_RGBA:	//could strip to PTI_BC1_RGB
		case PTI_BC2_RGBA_SRGB:	//could strip to PTI_BC1_RGB
		case PTI_BC3_RGBA:	//could strip to PTI_BC1_RGB
		case PTI_BC3_RGBA_SRGB:	//could strip to PTI_BC1_RGB
		case PTI_BC7_RGBA:	//much too messy...
		case PTI_BC7_RGBA_SRGB:
		case PTI_ETC2_RGB8A1: //would need to force the 'opaque' bit in each block and treat as PTI_ETC2_RGB8.
		case PTI_ETC2_RGB8A1_SRGB: //would need to force the 'opaque' bit in each block and treat as PTI_ETC2_RGB8.
		case PTI_ETC2_RGB8A8: //could strip to PTI_ETC2_RGB8
		case PTI_ETC2_RGB8A8_SRGB: //could strip to PTI_ETC2_SRGB8
		case PTI_ASTC_4X4:
		case PTI_ASTC_4X4_SRGB:
		case PTI_ASTC_5X4:
		case PTI_ASTC_5X4_SRGB:
		case PTI_ASTC_5X5:
		case PTI_ASTC_5X5_SRGB:
		case PTI_ASTC_6X5:
		case PTI_ASTC_6X5_SRGB:
		case PTI_ASTC_6X6:
		case PTI_ASTC_6X6_SRGB:
		case PTI_ASTC_8X5:
		case PTI_ASTC_8X5_SRGB:
		case PTI_ASTC_8X6:
		case PTI_ASTC_8X6_SRGB:
		case PTI_ASTC_10X5:
		case PTI_ASTC_10X5_SRGB:
		case PTI_ASTC_10X6:
		case PTI_ASTC_10X6_SRGB:
		case PTI_ASTC_8X8:
		case PTI_ASTC_8X8_SRGB:
		case PTI_ASTC_10X8:
		case PTI_ASTC_10X8_SRGB:
		case PTI_ASTC_10X10:
		case PTI_ASTC_10X10_SRGB:
		case PTI_ASTC_12X10:
		case PTI_ASTC_12X10_SRGB:
		case PTI_ASTC_12X12:
		case PTI_ASTC_12X12_SRGB:
#ifdef FTE_TARGET_WEB
		case PTI_WHOLEFILE:
#endif
			//erk. meh.
			break;
		case PTI_L8:
		case PTI_L8_SRGB:
		case PTI_R16:
		case PTI_P8:
		case PTI_R8:
		case PTI_R8_SNORM:
		case PTI_RG8:
		case PTI_RG8_SNORM:
		case PTI_RGB565:
		case PTI_RGB8:
		case PTI_BGR8:
		case PTI_E5BGR9:
		case PTI_RGBX8:
		case PTI_BGRX8:
		case PTI_RGBX8_SRGB:
		case PTI_BGRX8_SRGB:
		case PTI_BC1_RGB:
		case PTI_BC1_RGB_SRGB:
		case PTI_BC4_R8:
		case PTI_BC4_R8_SNORM:
		case PTI_BC5_RG8:
		case PTI_BC5_RG8_SNORM:
		case PTI_BC6_RGB_UFLOAT:
		case PTI_BC6_RGB_SFLOAT:
		case PTI_ETC1_RGB8:
		case PTI_ETC2_RGB8:
		case PTI_ETC2_RGB8_SRGB:
		case PTI_EAC_R11:
		case PTI_EAC_R11_SNORM:
		case PTI_EAC_RG11:
		case PTI_EAC_RG11_SNORM:
		case PTI_R16F:
		case PTI_R32F:
			break;	//already no alpha in these formats
		case PTI_DEPTH16:
		case PTI_DEPTH24:
		case PTI_DEPTH32:
		case PTI_DEPTH24_8:
			break;
		case PTI_EMULATED:
		case PTI_MAX: break;	//stfu
		}
		//FIXME: fill alpha channel with 255?
	}

	if ((vid.flags & VID_SRGBAWARE) /*&& (flags & IF_SRGB)*/ && !(flags & IF_NOSRGB))
	{	//most modern editors write srgb images.
		//however, that might not be supported.
		uploadfmt_t nf = PTI_MAX;
		switch(mips->encoding)
		{
		case PTI_L8:			nf = PTI_L8_SRGB; break;
		case PTI_L8A8:			nf = PTI_L8A8_SRGB; break;
		case PTI_LLLX8:			nf = PTI_RGBX8_SRGB; break;
		case PTI_LLLA8:			nf = PTI_RGBA8_SRGB; break;
		case PTI_RGBA8:			nf = PTI_RGBA8_SRGB; break;
		case PTI_RGBX8:			nf = PTI_RGBX8_SRGB; break;
		case PTI_BGRA8:			nf = PTI_BGRA8_SRGB; break;
		case PTI_BGRX8:			nf = PTI_BGRX8_SRGB; break;
		case PTI_BC1_RGB:		nf = PTI_BC1_RGB_SRGB; break;
		case PTI_BC1_RGBA:		nf = PTI_BC1_RGBA_SRGB; break;
		case PTI_BC2_RGBA:		nf = PTI_BC2_RGBA_SRGB; break;
		case PTI_BC3_RGBA:		nf = PTI_BC3_RGBA_SRGB; break;
		case PTI_BC7_RGBA:		nf = PTI_BC7_RGBA_SRGB; break;
		case PTI_ETC1_RGB8:		nf = PTI_ETC2_RGB8_SRGB; break;
		case PTI_ETC2_RGB8:		nf = PTI_ETC2_RGB8_SRGB; break;
		case PTI_ETC2_RGB8A1:	nf = PTI_ETC2_RGB8A1_SRGB; break;
		case PTI_ETC2_RGB8A8:	nf = PTI_ETC2_RGB8A8_SRGB; break;
		case PTI_ASTC_4X4:		nf = PTI_ASTC_4X4_SRGB; break;
		case PTI_ASTC_5X4:		nf = PTI_ASTC_5X4_SRGB; break;
		case PTI_ASTC_5X5:		nf = PTI_ASTC_5X5_SRGB; break;
		case PTI_ASTC_6X5:		nf = PTI_ASTC_6X5_SRGB; break;
		case PTI_ASTC_6X6:		nf = PTI_ASTC_6X6_SRGB; break;
		case PTI_ASTC_8X5:		nf = PTI_ASTC_8X5_SRGB; break;
		case PTI_ASTC_8X6:		nf = PTI_ASTC_8X6_SRGB; break;
		case PTI_ASTC_10X5:		nf = PTI_ASTC_10X5_SRGB; break;
		case PTI_ASTC_10X6:		nf = PTI_ASTC_10X6_SRGB; break;
		case PTI_ASTC_8X8:		nf = PTI_ASTC_8X8_SRGB; break;
		case PTI_ASTC_10X8:		nf = PTI_ASTC_10X8_SRGB; break;
		case PTI_ASTC_10X10:	nf = PTI_ASTC_10X10_SRGB; break;
		case PTI_ASTC_12X10:	nf = PTI_ASTC_12X10_SRGB; break;
		case PTI_ASTC_12X12:	nf = PTI_ASTC_12X12_SRGB; break;
		default:
			if (freedata)
				BZ_Free(rgbadata);
			return false;
		}
		if (sh_config.texfmt[nf])
			mips->encoding = nf;
		else
		{	//srgb->linear
			int m = mips->mip[0].width*mips->mip[0].height*mips->mip[0].depth;

			switch(nf)
			{
			case PTI_R8:
			case PTI_L8:
				for (i = 0; i < m; i++)
					((qbyte*)rgbadata)[i+0] = 255*Image_LinearFloatFromsRGBFloat(((qbyte*)rgbadata)[i+0] * (1.0/255));
				break;
			case PTI_L8A8:
				m*=2;
				for (i = 0; i < m; i+=2)
					((qbyte*)rgbadata)[i+0] = 255*Image_LinearFloatFromsRGBFloat(((qbyte*)rgbadata)[i+0] * (1.0/255));
				break;
			case PTI_R16:
				for (i = 0; i < m; i+=4)
					((unsigned short*)rgbadata)[i+0] = 0xffff*Image_LinearFloatFromsRGBFloat(((unsigned short*)rgbadata)[i+0] * (1.0/0xffff));
				break;
			case PTI_RGBA16:
				m*=4;
				for (i = 0; i < m; i+=4)
				{
					((unsigned short*)rgbadata)[i+0] = 0xffff*Image_LinearFloatFromsRGBFloat(((unsigned short*)rgbadata)[i+0] * (1.0/0xffff));
					((unsigned short*)rgbadata)[i+1] = 0xffff*Image_LinearFloatFromsRGBFloat(((unsigned short*)rgbadata)[i+1] * (1.0/0xffff));
					((unsigned short*)rgbadata)[i+2] = 0xffff*Image_LinearFloatFromsRGBFloat(((unsigned short*)rgbadata)[i+2] * (1.0/0xffff));
				}
				break;
			case PTI_RGBA8:
			case PTI_RGBX8:
			case PTI_BGRA8:
			case PTI_BGRX8:
				m*=4;
				for (i = 0; i < m; i+=4)
				{
					((qbyte*)rgbadata)[i+0] = 255*Image_LinearFloatFromsRGBFloat(((qbyte*)rgbadata)[i+0] * (1.0/255));
					((qbyte*)rgbadata)[i+1] = 255*Image_LinearFloatFromsRGBFloat(((qbyte*)rgbadata)[i+1] * (1.0/255));
					((qbyte*)rgbadata)[i+2] = 255*Image_LinearFloatFromsRGBFloat(((qbyte*)rgbadata)[i+2] * (1.0/255));
				}
				break;
			case PTI_BC1_RGB:
			case PTI_BC1_RGBA:
			case PTI_BC2_RGBA:
			case PTI_BC3_RGBA:
				//FIXME: bc1/2/3 has two leading 16bit 565 values per block.
			default:
				//these formats are weird. we can't just fiddle with the rgbdata
				//FIXME: etc2 has all sorts of weird encoding tables...
				if (freedata)
					BZ_Free(rgbadata);
				return false;
			}
		}
	}


	Image_RoundDimensions(&mips->mip[0].width, &mips->mip[0].height, flags);
	if (rgbadata)
	{
		if (mips->mip[0].width == imgwidth && mips->mip[0].height == imgheight)
			mips->mip[0].data = rgbadata;
		else
		{
			mips->mip[0].data = Image_ResampleTexture(mips->encoding, rgbadata, imgwidth, imgheight, NULL, mips->mip[0].width, mips->mip[0].height);
			if (mips->mip[0].data)
			{
				if (freedata)
					BZ_Free(rgbadata);
				freedata = true;
			}
			else
			{	//rescaling unsupported
				mips->mip[0].data = rgbadata;
				mips->mip[0].width = imgwidth;
				mips->mip[0].height = imgheight;
				mips->mip[0].depth = 1;
			}
		}
	}
	else
		mips->mip[0].data = NULL;
	Image_BlockSizeForEncoding(mips->encoding, &bb, &bw, &bh);
	mips->mip[0].datasize = ((mips->mip[0].width+bw-1)/bw) * ((mips->mip[0].height+bh-1)/bh) * bb;

	if (mips->type == PTI_3D)
	{
		qbyte *data2d = mips->mip[0].data, *data3d;
		mips->mip[0].data = NULL;
		/*our 2d input image is interlaced as y0z0,y0z1,y1z0,y1z1
		  however, hardware uses the more logical y0z0,y1z0,y0z1,y1z1 ordering (xis ordered properly already)*/
		if (mips->mip[0].height*mips->mip[0].height == mips->mip[0].width && mips->mip[0].depth == 1 && (mips->encoding == PTI_RGBA8 || mips->encoding == PTI_RGBX8 || mips->encoding == PTI_BGRA8 || mips->encoding == PTI_BGRX8))
		{
			int d, r;
			int size = mips->mip[0].height;
			mips->mip[0].width = size;
			mips->mip[0].height = size;
			mips->mip[0].depth = size;
			mips->mip[0].data = data3d = BZ_Malloc(size*size*size);
			for (d = 0; d < size; d++)
				for (r = 0; r < size; r++)
					memcpy(data3d + (r + d*size) * size, data2d + (r*size + d) * size, size*4);
			mips->mip[0].datasize = size*size*size*4;
		}
		if (freedata)
			BZ_Free(data2d);
		if (!mips->mip[0].data)
			return false;
	}

	if (flags & IF_PREMULTIPLYALPHA)
	{
		//works for rgba or bgra
		int i;
		switch(mips->encoding)
		{
		case PTI_RGBA32F:
			{
				float *fte_restrict premul = (float*)mips->mip[0].data;
				for (i = 0; i < mips->mip[0].width*mips->mip[0].height; i++, premul+=4)
				{
					premul[0] = (premul[0] * premul[3]);
					premul[1] = (premul[1] * premul[3]);
					premul[2] = (premul[2] * premul[3]);
				}
			}
			break;
		case PTI_RGBA16:
			{
				unsigned short *fte_restrict premul = (unsigned short*)mips->mip[0].data;
				for (i = 0; i < mips->mip[0].width*mips->mip[0].height; i++, premul+=4)
				{
					premul[0] = (premul[0] * premul[3])>>16;
					premul[1] = (premul[1] * premul[3])>>16;
					premul[2] = (premul[2] * premul[3])>>16;
				}
			}
			break;
		case PTI_LLLX8:	//FIXME: why the Xs?
		case PTI_LLLA8:
		case PTI_RGBA8:
		case PTI_RGBX8:
		case PTI_BGRA8:
		case PTI_BGRX8:
		case PTI_RGBA8_SRGB:	//fixme: what's the correct multiplication for srgb?
		case PTI_RGBX8_SRGB:
		case PTI_BGRA8_SRGB:
		case PTI_BGRX8_SRGB:
			{
				qbyte *fte_restrict premul = (qbyte*)mips->mip[0].data;
				for (i = 0; i < mips->mip[0].width*mips->mip[0].height; i++, premul+=4)
				{
					premul[0] = (premul[0] * premul[3])>>8;
					premul[1] = (premul[1] * premul[3])>>8;
					premul[2] = (premul[2] * premul[3])>>8;
				}
			}
			break;
		case PTI_L8A8:
		case PTI_L8A8_SRGB:
			{
				qbyte *fte_restrict premul = (qbyte*)mips->mip[0].data;
				for (i = 0; i < mips->mip[0].width*mips->mip[0].height; i++, premul+=2)
					premul[0] = (premul[0] * premul[1])>>8;
				break;
			}
		default:
			break;	//format not known, so no idea how to premultiply it. bc2/3 might already be premultiplied or not...
		}
	}

	mips->mip[0].needfree = freedata;
	return true;
}
//loads from a single mip. takes ownership of the data.
static qboolean Image_LoadRawTexture(texid_t tex, unsigned int flags, void *rawdata, void *palettedata, int imgwidth, int imgheight, uploadfmt_t fmt)
{
	struct pendingtextureinfo *mips;
	mips = Z_Malloc(sizeof(*mips));
	mips->type = (flags & IF_3DMAP)?PTI_3D:PTI_2D;

	if (!Image_GenMip0(mips, flags, rawdata, palettedata, imgwidth, imgheight, fmt, true))
	{
		Z_Free(mips);
		if (flags & IF_NOWORKER)
			Image_LoadTexture_Failed(tex, NULL, 0, 0);
		else
			COM_AddWork(WG_MAIN, Image_LoadTexture_Failed, tex, NULL, 0, 0);
		return false;
	}
	Image_GenerateMips(mips, flags);
	Image_ChangeFormat(mips, flags, fmt);

	tex->width = imgwidth;
	tex->height = imgheight;

	if (flags & IF_NOWORKER)
		Image_LoadTextureMips(tex, mips, 0, 0);
	else
		COM_AddWork(WG_MAIN, Image_LoadTextureMips, tex, mips, 0, 0);
	return true;
}

//always frees filedata, even on failure.
//also frees the textures fallback data, but only on success
static struct pendingtextureinfo *Image_LoadMipsFromMemory(int flags, const char *iname, const char *fname, qbyte *filedata, int filesize)
{
	uploadfmt_t format;
	qbyte *rgbadata;

	int imgwidth, imgheight;

	struct pendingtextureinfo *mips = NULL;

	//these formats have special handling, because they cannot be implemented via Read32BitImageFile - they don't result in rgba images.
#ifdef IMAGEFMT_KTX
	if (!mips)
		mips = Image_ReadKTXFile(flags, fname, filedata, filesize);
#endif
#ifdef IMAGEFMT_PKM
	if (!mips)
		mips = Image_ReadPKMFile(flags, fname, filedata, filesize);
#endif
#ifdef IMAGEFMT_DDS
	if (!mips)
		mips = Image_ReadDDSFile(flags, fname, filedata, filesize);
#endif
#ifdef IMAGEFMT_BLP
	if (!mips && filedata[0] == 'B' && filedata[1] == 'L' && filedata[2] == 'P' && filedata[3] == '2') 
		mips = Image_ReadBLPFile(flags, fname, filedata, filesize);
#endif
#ifdef IMAGEFMT_VTF
	if (!mips && filedata[0] == 'V' && filedata[1] == 'T' && filedata[2] == 'F' && filedata[3] == '\0')
		mips = Image_ReadVTFFile(flags, fname, filedata, filesize);
#endif

	//the above formats are assumed to have consumed filedata somehow (probably storing into mips->extradata)
	if (mips)
	{
		unsigned int picmip = min(Image_GetPicMip(flags), mips->mipcount-1), i;
		if (picmip < mips->mipcount)
		{
			for (i = 0; i < picmip; i++)
				if (mips->mip[i].needfree)
					BZ_Free(mips->mip[i].data);
			mips->mipcount -= i;
			memmove(mips->mip, mips->mip+i, sizeof(*mips->mip)*mips->mipcount);
		}

		Image_ChangeFormat(mips, flags, TF_INVALID);
		return mips;
	}

	if ((rgbadata = ReadRawImageFile(filedata, filesize, &imgwidth, &imgheight, &format, false, fname)))
	{
		extern cvar_t vid_hardwaregamma;
		if (!(flags&IF_NOGAMMA) && !vid_hardwaregamma.value)
			BoostGamma(rgbadata, imgwidth, imgheight, format);

		switch(format)
		{
		default:
			break;
		case PTI_RGBA32F:
		case PTI_RGBA16F:
		case PTI_L8A8:
		case PTI_RGBA8:
		case PTI_RGBA4444:
		case PTI_ARGB4444:
		case PTI_RGBA5551:
		case PTI_ARGB1555:
			flags &= ~IF_NOALPHA;
			break;
		case PTI_BGRX8_SRGB:
		case PTI_BGRX8:
		case PTI_RGBX8_SRGB:
		case PTI_RGBX8:
		case PTI_LLLX8:
			if (!(flags & IF_NOALPHA))
			{
				unsigned int alpha_width, alpha_height, p;
				char aname[MAX_QPATH];
				unsigned char *alphadata;
				char *alph;
				size_t alphsize;
				char ext[8];
				uploadfmt_t alphaformat;
				COM_StripExtension(fname, aname, sizeof(aname));
				COM_FileExtension(fname, ext, sizeof(ext));
				Q_strncatz(aname, "_alpha.", sizeof(aname));
				Q_strncatz(aname, ext, sizeof(aname));
				if (!strchr(aname, ':') && (alph = FS_LoadMallocFile (aname, &alphsize)))
				{
					if ((alphadata = ReadRawImageFile(alph, alphsize, &alpha_width, &alpha_height, &alphaformat, true, aname)))
					{
						if (alpha_width == imgwidth && alpha_height == imgheight)
						{
							for (p = 0; p < alpha_width*alpha_height; p++)
							{
								rgbadata[(p<<2) + 3] = (alphadata[(p<<2) + 0] + alphadata[(p<<2) + 1] + alphadata[(p<<2) + 2])/3;
							}
							switch(format)
							{
							case PTI_LLLX8:		format = PTI_RGBA8;		break;
							case PTI_RGBX8:		format = PTI_RGBA8;		break;
							case PTI_BGRX8:		format = PTI_BGRA8;		break;
							case PTI_RGBX8_SRGB:format = PTI_RGBA8_SRGB;break;
							case PTI_BGRX8_SRGB:format = PTI_BGRA8_SRGB;break;
							default:									break;
							}
						}
						BZ_Free(alphadata);
					}
					BZ_Free(alph);
				}
			}
			break;
		}

		mips = Z_Malloc(sizeof(*mips));
		mips->type = (flags & IF_3DMAP)?PTI_3D:PTI_2D;
		if (Image_GenMip0(mips, flags, rgbadata, NULL, imgwidth, imgheight, format, true))
		{
			Image_GenerateMips(mips, flags);
			Image_ChangeFormat(mips, flags, format);
			BZ_Free(filedata);
			return mips;
		}
		Z_Free(mips);
	}
#ifdef FTE_TARGET_WEB
	else if (1)
	{
		struct pendingtextureinfo *mips;
		mips = Z_Malloc(sizeof(*mips));
		mips->type = (flags & IF_3DMAP)?PTI_3D:PTI_2D;
		mips->mipcount = 1;
		mips->encoding = PTI_WHOLEFILE;
		mips->extrafree = NULL;
		mips->mip[0].width = 1;
		mips->mip[0].height = 1;
		mips->mip[0].depth = 1;
		mips->mip[0].data = filedata;
		mips->mip[0].datasize = filesize;
		mips->mip[0].needfree = true;
		//width+height are not yet known. bah.
		return mips;
	}
#endif
	else
		Con_Printf("Unable to read file %s (format unsupported)\n", fname);

	BZ_Free(filedata);
	return NULL;
}

//always frees filedata, even on failure.
//also frees the textures fallback data, but only on success
qboolean Image_LoadTextureFromMemory(texid_t tex, int flags, const char *iname, const char *fname, qbyte *filedata, int filesize)
{
	struct pendingtextureinfo *mips = Image_LoadMipsFromMemory(flags, iname, fname, filedata, filesize);
	if (mips)
	{
		BZ_Free(tex->fallbackdata);
		tex->fallbackdata = NULL;

		tex->width = mips->mip[0].width;
		tex->height = mips->mip[0].height;
		if ((flags & IF_NOWORKER) || Sys_IsMainThread())
			Image_LoadTextureMips(tex, mips, 0, 0);
		else
			COM_AddWork(WG_MAIN, Image_LoadTextureMips, tex, mips, 0, 0);
		return true;
	}
	return false;
}

static struct pendingtextureinfo *Image_LoadCubemapTextureData(const char *nicename, char *subpath, unsigned int texflags)
{
	static struct
	{
		const char *suffix;
		qboolean flipx, flipy, flipd;
		int pad;
	} cmscheme[][6] =
	{
		{
			{"rt", true,  false, true},
			{"lf", false, true,  true},
			{"ft", true,  true,  false},
			{"bk", false, false, false},
			{"up", true,  false, true},
			{"dn", true,  false, true}
		},

		{
			{"px", false, false, false},
			{"nx", false, false, false},
			{"py", false, false, false},
			{"ny", false, false, false},
			{"pz", false, false, false},
			{"nz", false, false, false}
		},

		{
			{"posx", false, false, false},
			{"negx", false, false, false},
			{"posy", false, false, false},
			{"negy", false, false, false},
			{"posz", false, false, false},
			{"negz", false, false, false}
		}
	};
	int i, j, e;
	struct pendingtextureinfo *mips;
	char fname[MAX_QPATH];
	size_t filesize;
	int width, height;
	uploadfmt_t format;
	char *nextprefix, *prefixend;
	size_t prefixlen;
	mips = Z_Malloc(sizeof(*mips));
	mips->type = PTI_CUBEMAP;
	mips->mipcount = 6;
	mips->encoding = PTI_RGBA8;
	mips->extrafree = NULL;

	for (i = 0; i < 6; i++)
	{
		prefixlen = 0;
		nextprefix = subpath;
		for(;;)
		{
			for (e = (texflags & IF_EXACTEXTENSION)?tex_extensions_count-1:0; e < tex_extensions_count; e++)
			{
				//try and open one
				qbyte *buf = NULL, *data;
				filesize = 0;

				for (j = 0; j < countof(cmscheme); j++)
				{
					Q_snprintfz(fname+prefixlen, sizeof(fname)-prefixlen, "%s_%s%s", nicename, cmscheme[j][i].suffix, tex_extensions[e].name);
					buf = FS_LoadMallocFile(fname, &filesize);
					if (buf)
						break;

					Q_snprintfz(fname+prefixlen, sizeof(fname)-prefixlen, "%s%s%s", nicename, cmscheme[j][i].suffix, tex_extensions[e].name);
					buf = FS_LoadMallocFile(fname, &filesize);
					if (buf)
						break;
				}

				//now read it
				if (buf)
				{
					if ((data = ReadRawImageFile(buf, filesize, &width, &height, &format, true, fname)))
					{
						extern cvar_t vid_hardwaregamma;
						if (width == height && (!i || width == mips->mip[0].width))	//cubemaps must be square and all the same size (npot is fine though)
						{	//(skies have a fallback for invalid sizes, but it'll run a bit slower)
							if (!(texflags&IF_NOGAMMA) && !vid_hardwaregamma.value)
								BoostGamma(data, width, height, format);
							mips->mip[i].data = R_FlipImage32(data, &width, &height, cmscheme[j][i].flipx, cmscheme[j][i].flipy, cmscheme[j][i].flipd);
							mips->mip[i].datasize = width*height*4;
							mips->mip[i].width = width;
							mips->mip[i].height = height;
							mips->mip[i].depth = 1;
							mips->mip[i].needfree = true;

							BZ_Free(buf);
							goto nextface;
						}
						BZ_Free(data);
					}
					BZ_Free(buf);
				}
			}

			//get ready for the next prefix...
			if (!nextprefix || !*nextprefix)
				break;	//no more...
			prefixend = strchr(nextprefix, ':');
			if (!prefixend)
				prefixend = nextprefix+strlen(nextprefix);

			prefixlen = prefixend-nextprefix;
			if (prefixlen >= sizeof(fname)-2)
				prefixlen = sizeof(fname)-2;
			memcpy(fname, nextprefix, prefixlen);
			fname[prefixlen++] = '/';

			if (*prefixend)
				prefixend++;
			nextprefix = prefixend;
		}

		while(i>0)
			BZ_Free(mips->mip[i--].data);
		Z_Free(mips);
		return NULL;
nextface:;
	}
	return mips;
}

qboolean Image_LocateHighResTexture(image_t *tex, flocation_t *bestloc, char *bestname, size_t bestnamesize, unsigned int *bestflags)
{
	char fname[MAX_QPATH], nicename[MAX_QPATH];
	int i, e;
	char *altname;
	char *nextalt;
	qboolean exactext = !!(tex->flags & IF_EXACTEXTENSION);

	int locflags = FSLF_DEPTH_INEXPLICIT|FSLF_DEEPONFAILURE;
	int bestdepth = 0x7fffffff, depth;
	int firstex = (tex->flags & IF_EXACTEXTENSION)?tex_extensions_count-1:0;

	flocation_t loc;
	
	if (strncmp(tex->ident, "http:", 5) && strncmp(tex->ident, "https:", 6))
	for(altname = tex->ident;altname;altname = nextalt)
	{
		nextalt = strchr(altname, ':');
		if (nextalt)
		{
			nextalt++;
			if (nextalt-altname >= sizeof(fname))
				continue;	//too long...
			memcpy(fname, altname, nextalt-altname-1);
			fname[nextalt-altname-1] = 0;
			altname = fname;
		}

		//see if we recognise the extension, and only strip it if we do.
		if (exactext)
			e = tex_extensions_count;
		else
		{
			COM_FileExtension(altname, nicename, sizeof(nicename));
			e = 0;
			if (strcmp(nicename, "lmp") && strcmp(nicename, "wal"))
				for (; e < tex_extensions_count; e++)
				{
					if (!strcmp(nicename, (*tex_extensions[e].name=='.')?tex_extensions[e].name+1:tex_extensions[e].name))
						break;
				}
		}

		//strip it and try replacements if we do, otherwise assume that we're meant to be loading progs/foo.mdl_0.tga or whatever
		if (e == tex_extensions_count || exactext)
		{
			exactext = true;
			Q_strncpyz(nicename, altname, sizeof(nicename));
		}
		else
			COM_StripExtension(altname, nicename, sizeof(nicename));

		if (!tex->fallbackdata || (gl_load24bit.ival && !(tex->flags & IF_NOREPLACE)))
		{
#ifdef IMAGEFMT_DDS
			Q_snprintfz(fname, sizeof(fname), "dds/%s.dds", nicename);
			depth = FS_FLocateFile(fname, locflags, &loc);
			if (depth < bestdepth)
			{
				Q_strncpyz(bestname, fname, bestnamesize);
				bestdepth = depth;
				*bestloc = loc;
				bestflags = 0;
			}
#endif

			if (strchr(nicename, '/') || strchr(nicename, '\\'))	//never look in a root dir for the pic
				i = 0;
			else
				i = 1;

			for (; i < sizeof(tex_path)/sizeof(tex_path[0]); i++)
			{
				if (!tex_path[i].enabled)
					continue;
				if (tex_path[i].args >= 3)
				{	//this is a path that needs subpaths
					char subpath[MAX_QPATH];
					char basename[MAX_QPATH];
					char *s, *n;
					if (!tex->subpath || !*nicename)
						continue;

					s = COM_SkipPath(nicename);
					n = basename;
					while (*s && (*s != '.'||exactext) && n < basename+sizeof(basename)-5)
						*n++ = *s++;
					s = strchr(s, '_');
					if (s)
					{
						while (*s && n < basename+sizeof(basename)-5)
							*n++ = *s++;
					}
					*n = 0;

					for(s = tex->subpath; s; s = n)
					{
						//subpath a:b:c tries multiple possible sub paths, for compatibility
						n = strchr(s, ':');
						if (n)
						{
							if (n-s >= sizeof(subpath))
								*subpath = 0;
							else
							{
								memcpy(subpath, s, n-s);
								subpath[n-s] = 0;
							}
							n++;
						}
						else
							Q_strncpyz(subpath, s, sizeof(subpath));
						for (e = firstex; e < tex_extensions_count; e++)
						{
							if (tex->flags & IF_NOPCX)
								if (!strcmp(tex_extensions[e].name, ".pcx"))
									continue;
							Q_snprintfz(fname, sizeof(fname), tex_path[i].path, subpath, basename, tex_extensions[e].name);
							depth = FS_FLocateFile(fname, locflags, &loc);
							if (depth < bestdepth)
							{
								Q_strncpyz(bestname, fname, bestnamesize);
								bestdepth = depth;
								*bestloc = loc;
								bestflags = 0;
							}
						}
					}
				}
				else
				{
					for (e = firstex; e < tex_extensions_count; e++)
					{
						if (tex->flags & IF_NOPCX)
							if (!strcmp(tex_extensions[e].name, ".pcx"))
								continue;
						Q_snprintfz(fname, sizeof(fname), tex_path[i].path, nicename, tex_extensions[e].name);
						depth = FS_FLocateFile(fname, locflags, &loc);
						if (depth < bestdepth)
						{
							Q_strncpyz(bestname, fname, bestnamesize);
							bestdepth = depth;
							*bestloc = loc;
							bestflags = 0;
						}
					}
				}

				//support expansion of _bump textures to _norm textures.
				if (tex->flags & IF_TRYBUMP)
				{
					if (tex_path[i].args >= 3)
					{
						/*no legacy compat needed, hopefully*/
					}
					else
					{
						char bumpname[MAX_QPATH], *b;
						const char *n;
						b = bumpname;
						n = nicename;
						while(*n)
						{
							if (*n == '_' && !strcmp(n, "_norm"))
							{
								strcpy(b, "_bump");
								b += 5;
								n += 5;
								break;
							}
							*b++ = *n++;
						}
						if (*n)	//no _norm, give up with that
							continue;
						*b = 0;
						for (e = firstex; e < tex_extensions_count; e++)
						{
							if (!strcmp(tex_extensions[e].name, ".tga"))
							{
								Q_snprintfz(fname, sizeof(fname), tex_path[i].path, bumpname, tex_extensions[e].name);

								Q_snprintfz(fname, sizeof(fname), tex_path[i].path, nicename, tex_extensions[e].name);
								depth = FS_FLocateFile(fname, locflags, &loc);
								if (depth < bestdepth)
								{
									Q_strncpyz(bestname, fname, bestnamesize);
									bestdepth = depth;
									*bestloc = loc;
									*bestflags = IF_TRYBUMP;
								}
							}
						}
					}
				}
			}


			/*still failed? attempt to load quake lmp files, which have no real format id (hence why they're not above)*/
			Q_strncpyz(fname, nicename, sizeof(fname));
			COM_DefaultExtension(fname, ".lmp", sizeof(fname));
			if (!(tex->flags & IF_NOPCX))
			{
				depth = FS_FLocateFile(fname, locflags, &loc);
				if (depth < bestdepth)
				{
					Q_strncpyz(bestname, fname, bestnamesize);
					bestdepth = depth;
					*bestloc = loc;
					bestflags = 0;
				}
			}
		}
	}

	return bestdepth != 0x7fffffff;
}

static void Image_LoadHiResTextureWorker(void *ctx, void *data, size_t a, size_t b)
{
	image_t *tex = ctx;
	char fname[MAX_QPATH];
	char fname2[MAX_QPATH];
	char *altname;
	char *nextalt;

	flocation_t loc;
	unsigned int locflags = 0;

	vfsfile_t *f;
	size_t fsize;
	char *buf;

	int i, j;
	int imgwidth;
	int imgheight;
	uploadfmt_t format;

	if ((tex->flags & IF_TEXTYPE) == IF_CUBEMAP)
	{	//cubemaps require special handling because they are (normally) 6 files instead of 1.
		//the exception is single-file dds cubemaps, but we don't support those.
		for(altname = tex->ident;altname;altname = nextalt)
		{
			struct pendingtextureinfo *mips = NULL;
			static const char *cubeexts[] =
			{
				""
#ifdef IMAGEFMT_KTX
				, ".ktx"
#endif
#ifdef IMAGEFMT_DDS
				, ".dds"
#endif
			};

			nextalt = strchr(altname, ':');
			if (nextalt)
			{
				nextalt++;
				if (nextalt-altname >= sizeof(fname))
					continue;	//too long...
				memcpy(fname, altname, nextalt-altname-1);
				fname[nextalt-altname-1] = 0;
				altname = fname;
			}

			for (i = 0; i < countof(tex_path) && !mips; i++)
			{
				if (!tex_path[i].enabled)
					continue;
				buf = NULL;
				fsize = 0;
				if (tex_path[i].args == 3 && tex->subpath)
				{
					char subpath[MAX_QPATH];
					char *n = tex->subpath, *e;
					while (*n)
					{
						e = strchr(n, ':');
						if (!e)
							e = n+strlen(n);
						Q_strncpyz(subpath, n, min(sizeof(subpath), (e-n)+1));
						n = e;
						while(*n == ':')
							n++;
						for (j = 0; !buf && j < countof(cubeexts); j++)
						{
							Q_snprintfz(fname2, sizeof(fname2), tex_path[i].path, subpath, altname, cubeexts[j]);
							buf = FS_LoadMallocFile(fname2, &fsize);
						}
					}
				}
				else if (tex_path[i].args == 2)
				{
					for (j = 0; !buf && j < countof(cubeexts); j++)
					{
						Q_snprintfz(fname2, sizeof(fname2), tex_path[i].path, altname, cubeexts[j]);
						buf = FS_LoadMallocFile(fname2, &fsize);
					}
				}
				if (buf)
				{
#ifdef IMAGEFMT_KTX
					if (!mips)
						mips = Image_ReadKTXFile(tex->flags, altname, buf, fsize);
#endif
#ifdef IMAGEFMT_DDS
					if (!mips)
						mips = Image_ReadDDSFile(tex->flags, altname, buf, fsize);
#endif
					if (!mips)
						BZ_Free(buf);
				}
			}

			if (!mips)	//try to load multiple images
				mips = Image_LoadCubemapTextureData(altname, tex->subpath, tex->flags);

			if (mips)
			{
				tex->width = mips->mip[0].width;
				tex->height = mips->mip[0].height;

				if (tex->flags & IF_NOWORKER)
					Image_LoadTextureMips(tex, mips, 0, 0);
				else
					COM_AddWork(WG_MAIN, Image_LoadTextureMips, tex, mips, 0, 0);
				return;
			}
		}
		if (tex->flags & IF_NOWORKER)
			Image_LoadTexture_Failed(tex, NULL, 0, 0);
		else
			COM_AddWork(WG_MAIN, Image_LoadTexture_Failed, tex, NULL, 0, 0);
		return;
	}

	

	if (Image_LocateHighResTexture(tex, &loc, fname, sizeof(fname), &locflags))
	{
		f = FS_OpenReadLocation(&loc);
		if (f)
		{
			fsize = VFS_GETLEN(f);
			buf = BZ_Malloc(fsize);
			if (buf)
			{
				VFS_READ(f, buf, fsize);
				VFS_CLOSE(f);

				if (locflags & IF_TRYBUMP)
				{	//it was supposed to be a heightmap image (that we need to convert to normalmap)
					qbyte *d;
					int w, h;
					uploadfmt_t fmt;
					if ((d = ReadTargaFile(buf, fsize, &w, &h, &fmt, true, PTI_L8)))	//Only load a greyscale image.
					{
						BZ_Free(buf);
						if (Image_LoadRawTexture(tex, tex->flags, d, NULL, w, h, TF_HEIGHT8))
						{
							BZ_Free(tex->fallbackdata);
							tex->fallbackdata = NULL;	
							return;
						}
					}
					//guess not, fall back to normalmaps
				}

				if (Image_LoadTextureFromMemory(tex, tex->flags, tex->ident, fname, buf, fsize))
				{
					BZ_Free(tex->fallbackdata);
					tex->fallbackdata = NULL;
					return;
				}
			}
			else
				VFS_CLOSE(f);
		}
	}

	//now look in wad files and swap over the fallback. (halflife compatability)
	COM_StripExtension(tex->ident, fname, sizeof(fname));
	buf = W_GetTexture(fname, &imgwidth, &imgheight, &format);
	if (buf)
	{
		BZ_Free(tex->fallbackdata);
		tex->fallbackdata = buf;
		tex->fallbackfmt = format;
		tex->fallbackwidth = imgwidth;
		tex->fallbackheight = imgheight;
	}

	if (tex->fallbackdata)
	{
		if (tex->fallbackfmt == TF_INVALID)
		{
			void *data = tex->fallbackdata;
			tex->fallbackdata = NULL;
			if (Image_LoadTextureFromMemory(tex, tex->flags, tex->ident, fname, data, tex->fallbackwidth))
				return;
		}
		else if (Image_LoadRawTexture(tex, tex->flags, tex->fallbackdata, (char*)tex->fallbackdata+(tex->fallbackwidth*tex->fallbackheight), tex->fallbackwidth, tex->fallbackheight, tex->fallbackfmt))
		{
			tex->fallbackdata = NULL;
			return;
		}
		tex->fallbackdata = NULL;	
	}

//	Sys_Printf("Texture %s failed\n", nicename);
	//signal the main thread to set the final status instead of just setting it to avoid deadlock (it might already be waiting for it).
	if (tex->flags & IF_NOWORKER)
		Image_LoadTexture_Failed(tex, NULL, 0, 0);
	else
		COM_AddWork(WG_MAIN, Image_LoadTexture_Failed, tex, NULL, 0, 0);
}

//returns the pointer if its valid, otherwise null
//this is to pass pointers via the console
image_t *Image_TextureIsValid(qintptr_t address)
{
	image_t *img;
	for (img = imagelist; img; img = img->next)
	{
		if (img == (image_t*)address)
			break;
	}
	return img;
}

//find an existing texture
image_t *Image_FindTexture(const char *identifier, const char *subdir, unsigned int flags)
{
	image_t *tex;
	if (!subdir)
		subdir = "";
	tex = Hash_Get(&imagetable, identifier);
	while(tex)
	{
		if (!((tex->flags ^ flags) & (IF_CLAMP|IF_PALETTIZE|IF_PREMULTIPLYALPHA)))
		{
#ifdef PURGEIMAGES
			if (!strcmp(subdir, tex->subpath?tex->subpath:""))
#endif
			{
				tex->regsequence = r_regsequence;
				return tex;
			}
		}
		tex = Hash_GetNext(&imagetable, identifier, tex);
	}
	return NULL;
}
//create a texture, with dupes. you'll need to load something into it too.
static image_t *Image_CreateTexture_Internal (const char *identifier, const char *subdir, unsigned int flags)
{
	image_t *tex;
	bucket_t *buck;

	tex = Z_Malloc(sizeof(*tex) + sizeof(bucket_t) + strlen(identifier)+1 + (subdir?strlen(subdir)+1:0));
	buck = (bucket_t*)(tex+1);
	tex->ident = (char*)(buck+1);
	strcpy(tex->ident, identifier);
#ifdef _DEBUG
	Q_strncpyz(tex->dbgident, identifier, sizeof(tex->dbgident));
#endif
	if (subdir && *subdir)
	{
		tex->subpath = tex->ident + strlen(identifier)+1;
		strcpy(tex->subpath, subdir);
	}

	tex->next = imagelist;
	imagelist = tex;

	if ((vid.flags & VID_SRGBAWARE) && !(flags & IF_NOSRGB))
		tex->flags = flags | IF_SRGB;	//guess...
	else
		tex->flags = flags;
	tex->width = 0;
	tex->height = 0;
	tex->regsequence = r_regsequence;
	tex->status = TEX_NOTLOADED;
	tex->fallbackdata = NULL;
	tex->fallbackwidth = 0;
	tex->fallbackheight = 0;
	tex->fallbackfmt = TF_INVALID;
	if (*tex->ident)
		Hash_Add(&imagetable, tex->ident, tex, buck);
	return tex;
}

image_t *Image_CreateTexture (const char *identifier, const char *subdir, unsigned int flags)
{
	image_t *image;
#ifdef LOADERTHREAD
	Sys_LockMutex(com_resourcemutex);
#endif
	image = Image_CreateTexture_Internal(identifier, subdir, flags);
#ifdef LOADERTHREAD
	Sys_UnlockMutex(com_resourcemutex);
#endif
	return image;
}

#ifdef WEBCLIENT
//called on main thread. oh well.
static void Image_Downloaded(struct dl_download *dl)
{
	qboolean success = false;
	image_t *tex = dl->user_ctx;
	image_t *p;

	//make sure the renderer wasn't restarted mid-download
	for (p = imagelist; p; p = p->next)
		if (p == tex)
			break;
	if (p)
	{
		if (dl->status == DL_FINISHED)
		{
			size_t fsize = VFS_GETLEN(dl->file);
			char *buf = BZ_Malloc(fsize);
			if (VFS_READ(dl->file, buf, fsize) == fsize)
				if (Image_LoadTextureFromMemory(tex, tex->flags, tex->ident, dl->url, buf, fsize))
					success = true;
		}
		if (!success)
			Image_LoadTexture_Failed(tex, NULL, 0, 0);
	}
}
#endif

//find a texture. will try to load it from disk, using the fallback if it would fail.
image_t *QDECL Image_GetTexture(const char *identifier, const char *subpath, unsigned int flags, void *fallbackdata, void *fallbackpalette, int fallbackwidth, int fallbackheight, uploadfmt_t fallbackfmt)
{
	image_t *tex;

	qboolean dontposttoworker = (flags & (IF_NOWORKER | IF_LOADNOW));
	qboolean lowpri = (flags & IF_LOWPRIORITY);
	qboolean highpri = (flags & IF_HIGHPRIORITY);
	flags &= ~(IF_LOADNOW | IF_LOWPRIORITY | IF_HIGHPRIORITY);

#ifdef LOADERTHREAD
	Sys_LockMutex(com_resourcemutex);
#endif
	tex = Image_FindTexture(identifier, subpath, flags);
	if (tex)
	{
		//FIXME: race condition is possible here
		//if a non-replaced texture is given a fallback while a non-fallback version is still loading, it can still fail.
		if (tex->status == TEX_FAILED && fallbackdata)
			tex->status = TEX_LOADING;
		else if (tex->status != TEX_NOTLOADED)
		{
#ifdef LOADERTHREAD
			Sys_UnlockMutex(com_resourcemutex);
#endif
			return tex;	//already exists
		}
		tex->flags = flags;
	}
	else
		tex = Image_CreateTexture_Internal(identifier, subpath, flags);

	tex->status = TEX_LOADING;
	if (fallbackdata)
	{
		int b = fallbackwidth*fallbackheight, pb = 0;
		switch(fallbackfmt)
		{
		case TF_INVALID:
			b = fallbackwidth;
			pb = fallbackheight;
			break;
		case TF_8PAL24:
			pb = 3*256;
			b *= 1;
			break;
		case TF_8PAL32:
			pb = 4*256;
			b *= 1;
			break;
		case PTI_R8:
		case PTI_P8:
		case TF_SOLID8:
		case TF_TRANS8:
		case TF_TRANS8_FULLBRIGHT:
		case TF_H2_T7G1:
		case TF_H2_TRANS8_0:
		case TF_H2_T4A4:
		case TF_HEIGHT8:
		case TF_HEIGHT8PAL:	//we don't care about the actual palette.
			b *= 1;
			break;
//		case PTI_LLLX8:
//		case PTI_LLLA8:
		case TF_RGBX32:
		case TF_RGBA32:
		case TF_BGRX32:
		case TF_BGRA32:
			b *= 4;
			break;
		case TF_MIP4_8PAL24:
		case TF_MIP4_8PAL24_T255:
			pb = 3*256;
		case TF_MIP4_P8:
		case TF_MIP4_SOLID8:
			b = (fallbackwidth>>0)*(fallbackheight>>0) +
				(fallbackwidth>>1)*(fallbackheight>>1) +
				(fallbackwidth>>2)*(fallbackheight>>2) +
				(fallbackwidth>>3)*(fallbackheight>>3);
			break;
		default:
			Sys_Error("Image_GetTexture: bad format");
		}
		tex->fallbackdata = BZ_Malloc(b + pb);
		memcpy(tex->fallbackdata, fallbackdata, b);
		if (pb)
			memcpy((qbyte*)tex->fallbackdata + b, fallbackpalette, pb);
		tex->fallbackwidth = fallbackwidth;
		tex->fallbackheight = fallbackheight;
		tex->fallbackfmt = fallbackfmt;
	}
	else
	{
		tex->fallbackdata = NULL;
		tex->fallbackwidth = 0;
		tex->fallbackheight = 0;
		tex->fallbackfmt = TF_INVALID;
	}
#ifdef LOADERTHREAD
	Sys_UnlockMutex(com_resourcemutex);
#endif
	//FIXME: pass fallback through this way instead?

	if (dontposttoworker)
		Image_LoadHiResTextureWorker(tex, NULL, 0, 0);
	else
	{
#ifdef WEBCLIENT
		if (!strncmp(tex->ident, "http://", 7) || !strncmp(tex->ident, "https://", 8))
		{
			struct dl_download *dl;
			size_t sizelimit = max(0,r_image_downloadsizelimit.ival);
			if (sizelimit>0 || !*r_image_downloadsizelimit.string)
				dl = HTTP_CL_Get(tex->ident, NULL, Image_Downloaded);
			else
			{
				Con_Printf("r_image_downloadsizelimit: image downloading is blocked\n");
				dl = NULL;
			}
			if (dl)
			{
				if (sizelimit)
					dl->sizelimit = sizelimit;
				dl->user_ctx = tex;
				dl->file = VFSPIPE_Open(1, false);
				dl->isquery = true;
			}
#ifdef MULTITHREAD
			DL_CreateThread(dl, NULL, NULL);
#else
			tex->status = TEX_FAILED;	//HACK: so nothing waits for it.
#endif
		}
		else
#endif
			if (highpri)
			COM_InsertWork(WG_LOADER, Image_LoadHiResTextureWorker, tex, NULL, 0, 0);
		else if (lowpri)
			COM_AddWork(WG_LOADER, Image_LoadHiResTextureWorker, tex, NULL, 0, 0);
		else
			COM_AddWork(WG_LOADER, Image_LoadHiResTextureWorker, tex, NULL, 0, 0);
	}
	return tex;
}
void Image_Upload			(texid_t tex, uploadfmt_t fmt, void *data, void *palette, int width, int height, unsigned int flags)
{
	struct pendingtextureinfo mips;
	size_t i;

	//skip if we're not actually changing the data/size/format.
	if (!data && tex->format == fmt && tex->width == width && tex->height == height  && tex->depth == 1)
		return;

	mips.extrafree = NULL;
	mips.type = (flags & IF_3DMAP)?PTI_3D:PTI_2D;
	if (!Image_GenMip0(&mips, flags, data, palette, width, height, fmt, false))
		return;
	Image_GenerateMips(&mips, flags);
	Image_ChangeFormat(&mips, flags, fmt);
	rf->IMG_LoadTextureMips(tex, &mips);
	tex->format = fmt;
	tex->width = width;
	tex->height = height;
	tex->depth = 1;
	tex->status = TEX_LOADED;

	for (i = 0; i < mips.mipcount; i++)
		if (mips.mip[i].needfree)
			BZ_Free(mips.mip[i].data);
	if (mips.extrafree)
		BZ_Free(mips.extrafree);
}

typedef struct
{
	char *name;
	char *legacyname;
	int	maximize, minmip, minimize;
	int pad;
} texmode_t;
static texmode_t texmodes[] = {
	{"n",	"GL_NEAREST",					0,	-1,	0},
	{"l",	"GL_LINEAR",					1,	-1,	1},
	{"nn",	"GL_NEAREST_MIPMAP_NEAREST",	0,	0,	0},
	{"ln",	"GL_LINEAR_MIPMAP_NEAREST",		1,	0,	1},
	{"nl",	"GL_NEAREST_MIPMAP_LINEAR",		0,	1,	0},
	{"ll",	"GL_LINEAR_MIPMAP_LINEAR",		1,	1,	1},

	//more explicit names
	{"n.n",	NULL,							0,	-1,	0},
	{"l.l",	NULL,							1,	-1,	1},
	{"nnn",	NULL,							0,	0,	0},
	{"lnl",	NULL,							1,	0,	1},
	{"nln",	NULL,							0,	1,	0},
	{"lll",	NULL,							1,	1,	1},

	//inverted mag filters
	{"n.l",	NULL,							0,	-1,	1},
	{"l.n",	NULL,							1,	-1,	0},
	{"nnl",	NULL,							0,	0,	1},
	{"lnn",	NULL,							1,	0,	0},
	{"nll",	NULL,							0,	1,	1},
	{"lln",	NULL,							1,	1,	0}
};
static void Image_ParseTextureMode(char *cvarname, char *modename, int modes[3])
{
	int i;
	modes[0] = 1;
	modes[1] = 0;
	modes[2] = 1;
	for (i = 0; i < sizeof(texmodes) / sizeof(texmodes[0]); i++)
	{
		if (!Q_strcasecmp(modename, texmodes[i].name) || (texmodes[i].legacyname && !Q_strcasecmp(modename, texmodes[i].legacyname)))
		{
			modes[0] = texmodes[i].minimize;
			modes[1] = texmodes[i].minmip;
			modes[2] = texmodes[i].maximize;
			return;
		}
	}
	Con_Printf("%s: mode %s was not recognised\n", cvarname, modename);
}
void QDECL Image_TextureMode_Callback (struct cvar_s *var, char *oldvalue)
{
	int mip[3]={1,0,1}, pic[3]={1,-1,1}, mipcap[2] = {0, 1000};
	float anis = 1, lodbias = 0;
	char *s;
	extern cvar_t gl_texturemode, gl_texturemode2d, gl_texture_anisotropic_filtering, gl_texture_lodbias, gl_mipcap;

	Image_ParseTextureMode(gl_texturemode.name, gl_texturemode.string, mip);
	Image_ParseTextureMode(gl_texturemode2d.name, gl_texturemode2d.string, pic);
	anis = gl_texture_anisotropic_filtering.value;
	//parse d_mipcap (two values, nearest furthest)
	s = COM_Parse(gl_mipcap.string);
	mipcap[0] = *com_token?atoi(com_token):0;
//	if (mipcap[0] > 3)	/*cap it to 3, so no 16*16 textures get bugged*/
//		mipcap[0] = 3;
	s = COM_Parse(s);
	mipcap[1] = *com_token?atoi(com_token):1000;
	if (mipcap[1] < mipcap[0])
		mipcap[1] = mipcap[0];
	lodbias = gl_texture_lodbias.value;

	if (rf && rf->IMG_UpdateFiltering)
		rf->IMG_UpdateFiltering(imagelist, mip, pic, mipcap, lodbias, anis);
}
qboolean Image_UnloadTexture(image_t *tex)
{
	if (tex->status == TEX_LOADED)
	{
		rf->IMG_DestroyTexture(tex);
		tex->status = TEX_NOTLOADED;
		return true;
	}
	return false;
}

//nukes an existing texture, destroying all traces. any lingering references will cause problems, so be careful about how you access these.
void Image_DestroyTexture(image_t *tex)
{
	image_t **link;
	if (!tex)
		return;
	TEXDOWAIT(tex);	//just in case.
#ifdef LOADERTHREAD
	Sys_LockMutex(com_resourcemutex);
#endif
	Image_UnloadTexture(tex);

	for (link = &imagelist; *link; link = &(*link)->next)
	{
		if (*link == tex)
		{
			*link = tex->next;
			break;
		}
	}
#ifdef LOADERTHREAD
	Sys_UnlockMutex(com_resourcemutex);
#endif
	if (*tex->ident)
		Hash_RemoveData(&imagetable, tex->ident, tex);
	Z_Free(tex);
}


void Shader_TouchTextures(void);
void Image_Purge(void)
{
#ifdef PURGEIMAGES
	image_t *tex, *a;
	int loaded = 0, total = 0;
	size_t mem = 0;
	Shader_TouchTextures();
	for (tex = imagelist; tex; tex = tex->next)
	{
		if (tex->flags & IF_NOPURGE)
			continue;
		if (tex->regsequence != r_regsequence)
			Image_UnloadTexture(tex);
	}
#endif
}


void Image_List_f(void)
{
	flocation_t loc;
	image_t *tex, *a;
	int loaded = 0, aliases = 0, failed = 0, total = 0;
	size_t drivermem = 0;
	size_t aliasedmem = 0;
	size_t imgmem;
	unsigned int loadflags;
	char fname[MAX_QPATH];
	const char *filter = Cmd_Argv(1);
	for (tex = imagelist; tex; tex = tex->next)
	{
		total++;
		if (*filter && !strstr(tex->ident, filter))
			continue;
		if (tex->status == TEX_FAILED && !developer.ival)
		{	//don't show all the failures.
			failed++;
			continue;
		}
		if (tex->subpath)
			Con_Printf("^h(%s)^h", tex->subpath);
		Con_DLPrintf(1, " %x", tex->flags);
		
		if (Image_LocateHighResTexture(tex, &loc, fname, sizeof(fname), &loadflags))
		{
			char defuck[MAX_OSPATH], *bullshit;
			Q_strncpyz(defuck, loc.search->logicalpath, sizeof(defuck));
			while((bullshit=strchr(defuck, '\\')))
				*bullshit = '/';

//			if ((tex->flags&(IF_CLAMP|IF_PALETTIZE)) == 0)
				Con_Printf("^[%s\\tip\\%s/%s\\tipimgptr\\%#"PRIxSIZE"^]: ", tex->ident, defuck, fname, (size_t)tex);
//			else
//				Con_Printf("^[%s\\tip\\%s/%s^]: ", tex->ident, defuck, fname);
		}
		else
		{
			loc.len = 0;
			Con_Printf("^[%s\\tipimgptr\\%#"PRIxSIZE"^]: ", tex->ident, (size_t)tex);
		}

		for (a = tex->aliasof; a; a = a->aliasof)
		{
			if (a->subpath)
				Con_Printf("^3^h(%s)^h%s: ", a->subpath, a->ident);
			else
				Con_Printf("^3%s: ", a->ident);
		}

		if (tex->status == TEX_LOADED)
		{
			unsigned int blockbytes, blockwidth, blockheight;
			Image_BlockSizeForEncoding(tex->format, &blockbytes, &blockwidth, &blockheight);
			imgmem = blockbytes * (tex->width+blockwidth-1)/blockwidth * (tex->height+blockheight-1)/blockheight;
			if (!(tex->flags & IF_NOMIPMAP))
				imgmem += imgmem/3;	//mips take about a third extra mem.
			Con_Printf("^2loaded (%i*%i ^4%s^2, %3fKB->%3fKB)\n", tex->width, tex->height, Image_FormatName(tex->format), loc.len/(1024.0), imgmem/(1024.0));
			if (tex->aliasof)
			{
				aliasedmem += imgmem;
				aliases++;
			}
			else
			{
				drivermem += imgmem;
				loaded++;
			}
		}
		else if (tex->status == TEX_FAILED)
		{
			Con_Printf("^1failed\n");
			failed++;
		}
		else if (tex->status == TEX_NOTLOADED)
			Con_Printf("^5not loaded\n");
		else
			Con_Printf("^bloading\n");
	}

	Con_Printf("%i images total\n", total);
	Con_Printf("%i images loaded (%gMB)\n", loaded, drivermem/(1024*1024.0));
	Con_Printf("%i image alises (%gMB)\n", aliases, aliasedmem/(1024*1024.0));
	Con_Printf("%i images failed\n", failed);
}

void Image_Formats_f(void)
{
	size_t i;

#ifdef GLQUAKE
	if (qrenderer == QR_OPENGL)
	{
		Con_Printf("OpenGL info:\n");
		Con_Printf("OpenGL Version: %s%g\n", gl_config.gles?"ES ":"", gl_config.glversion);
		Con_Printf("OpenGLSL Version: %i\n", gl_config.maxglslversion);
		Con_Printf("OpenGLSL Attributes: %u\n", gl_config.maxattribs);

		Con_Printf("arb_texture_env_combine: %u\n", gl_config.arb_texture_env_combine);
		Con_Printf("arb_texture_env_dot3: %u\n", gl_config.arb_texture_env_dot3);
		Con_Printf("arb_texture_compression: %u\n", gl_config.arb_texture_compression);
		Con_Printf("geometryshaders: %u\n", gl_config.geometryshaders);
		Con_Printf("ext_framebuffer_objects: %u\n", gl_config.ext_framebuffer_objects);
		Con_Printf("arb_framebuffer_srgb: %u\n", gl_config.arb_framebuffer_srgb);
		Con_Printf("arb_shader_objects: %u\n", gl_config.arb_shader_objects);
		Con_Printf("arb_shadow: %u\n", gl_config.arb_shadow);
		Con_Printf("arb_depth_texture: %u\n", gl_config.arb_depth_texture);
		Con_Printf("ext_stencil_wrap: %u\n", gl_config.ext_stencil_wrap);
		Con_Printf("ext_packed_depth_stencil: %u\n", gl_config.ext_packed_depth_stencil);
		Con_Printf("arb_depth_clamp: %u\n", gl_config.arb_depth_clamp);
		Con_Printf("ext_texture_filter_anisotropic: %u\n", gl_config.ext_texture_filter_anisotropic);
	}
#endif

	Con_Printf(		"            Programs: "S_COLOR_GREEN"%s\n", sh_config.progs_supported?va(sh_config.progpath, "*"):S_COLOR_RED"Unsupported");
	if (sh_config.progs_supported)
	{
		Con_Printf(	"     Shader versions: %u - %u\n", sh_config.minver, sh_config.maxver);
		Con_Printf(	"       Max GPU Bones: %s%u\n", sh_config.max_gpu_bones?S_COLOR_GREEN:S_COLOR_RED, sh_config.max_gpu_bones);
	}
	Con_Printf(		"     Legacy Pipeline: %s\n", sh_config.progs_required?S_COLOR_RED"Unsupported":S_COLOR_GREEN"Supported");
	if (!sh_config.progs_required)
	{
		Con_Printf(	"       Env_Combiners: %s\n", sh_config.nv_tex_env_combine4?S_COLOR_GREEN"Extended":sh_config.tex_env_combine?S_COLOR_GREEN"Supported":S_COLOR_RED"Unsupported");
		Con_Printf(	"             Env_Add: %s\n", sh_config.env_add?S_COLOR_GREEN"Supported":S_COLOR_RED"Unsupported");
	}
	Con_Printf(		"  Max Texture2d Size: %s%u*%u\n", S_COLOR_GREEN, sh_config.texture2d_maxsize, sh_config.texture2d_maxsize);
	Con_Printf(		"Max Texture2d Layers: %s%u\n", sh_config.texture2darray_maxlayers?S_COLOR_GREEN:S_COLOR_RED, sh_config.texture2darray_maxlayers);
	Con_Printf(		"  Max Texture3d Size: %s%u*%u*%u\n", sh_config.texture3d_maxsize?S_COLOR_GREEN:S_COLOR_RED, sh_config.texture3d_maxsize, sh_config.texture3d_maxsize, sh_config.texture3d_maxsize);
	Con_Printf(		"Max TextureCube Size: %s%u*%u\n", sh_config.havecubemaps?S_COLOR_GREEN:S_COLOR_RED, sh_config.texturecube_maxsize, sh_config.texturecube_maxsize);
	Con_Printf(		"    Non-Power-Of-Two: %s%s\n", sh_config.texture_non_power_of_two?S_COLOR_GREEN"Supported":(sh_config.texture_non_power_of_two_pic?S_COLOR_YELLOW"Limited":S_COLOR_RED"Unsupported"), sh_config.npot_rounddown?" (rounded down)":"");
	Con_Printf(		"  Block Size Padding: %s\n", sh_config.texture_allow_block_padding?S_COLOR_GREEN"Supported":S_COLOR_RED"Unsupported");
	Con_Printf(		"              Mipcap: %s\n", sh_config.can_mipcap?S_COLOR_GREEN"Supported":S_COLOR_RED"Unsupported");
	for (i = 0; i < PTI_MAX; i++)
	{
		switch(i)
		{
		case PTI_EMULATED:
			continue;
		default:
			break;
		}
		Con_Printf("%20s: %s\n", Image_FormatName(i), sh_config.texfmt[i]?S_COLOR_GREEN"Enabled":S_COLOR_RED"Disabled");
	}
}

//may not create any images yet.
void Image_Init(void)
{
	wadmutex = Sys_CreateMutex();
	memset(imagetablebuckets, 0, sizeof(imagetablebuckets));
	Hash_InitTable(&imagetable, sizeof(imagetablebuckets)/sizeof(imagetablebuckets[0]), imagetablebuckets);

	Cmd_AddCommandD("r_imagelist", Image_List_f, "Prints out a list of the currently-known textures.");
	Cmd_AddCommandD("r_imageformats", Image_Formats_f, "Prints out a list of the usable hardware pixel formats.");
}
//destroys all textures
void Image_Shutdown(void)
{
	image_t *tex;
	int i = 0, j = 0;
	Cmd_RemoveCommand("r_imagelist");
	while (imagelist)
	{
		tex = imagelist;
		if (*tex->ident)
			Hash_RemoveData(&imagetable, tex->ident, tex);
		imagelist = tex->next;
		if (tex->status == TEX_LOADED)
			j++;
		rf->IMG_DestroyTexture(tex);
		Z_Free(tex);
		i++;
	}
	if (i)
		Con_DPrintf("Destroyed %i/%i images\n", j, i);

	if (wadmutex)
		Sys_DestroyMutex(wadmutex);
	wadmutex = NULL;
}

//load the named file, without failing.
texid_t R_LoadHiResTexture(const char *name, const char *subpath, unsigned int flags)
{
	char nicename[MAX_QPATH], *data;
	if (!*name)
		return r_nulltex;
	Q_strncpyz(nicename, name, sizeof(nicename));
	while((data = strchr(nicename, '*')))
		*data = '#';
	return Image_GetTexture(nicename, subpath, flags, NULL, NULL, 0, 0, TF_INVALID);	//queues the texture creation.
}

//attempt to load the named texture
//will not load external textures if gl_load24bit is set (failing instantly if its just going to fail later on anyway)
//the specified data will be used if the high-res image is blocked/not found.
texid_t R_LoadReplacementTexture(const char *name, const char *subpath, unsigned int flags, void *lowres, int lowreswidth, int lowresheight, uploadfmt_t format)
{
	char nicename[MAX_QPATH], *data;
	if (!*name)
		return r_nulltex;
	if (!gl_load24bit.ival && !lowres)
		return r_nulltex;
	Q_strncpyz(nicename, name, sizeof(nicename));
	while((data = strchr(nicename, '*')))
		*data = '#';
	return Image_GetTexture(nicename, subpath, flags, lowres, NULL, lowreswidth, lowresheight, format);	//queues the texture creation.
}
#ifdef RTLIGHTS
void R_LoadNumberedLightTexture(dlight_t *dl, int cubetexnum)
{
	Q_snprintfz(dl->cubemapname, sizeof(dl->cubemapname), "cubemaps/%i", cubetexnum);
	if (!gl_load24bit.ival)
		dl->cubetexture = r_nulltex;
	else
		dl->cubetexture = Image_GetTexture(dl->cubemapname, NULL, IF_CUBEMAP, NULL, NULL, 0, 0, TF_INVALID);
}
#endif

#if 0
extern cvar_t r_shadow_bumpscale_bumpmap;
texid_t R_LoadBumpmapTexture(const char *name, const char *subpath)
{
	char *buf, *data;
	texid_t tex;
//	int h;
	char fname[MAX_QPATH], nicename[MAX_QPATH];
	qboolean hasalpha;

	static char *extensions[] =
	{//reverse order of preference - (match commas with optional file types)
		".tga",
		""
	};

	int i, e;

	TRACE(("dbg: Mod_LoadBumpmapTexture: texture %s\n", name));

	COM_StripExtension(name, nicename, sizeof(nicename));

	tex = R_FindTexture(name, 0);
	if (TEXVALID(tex))	//don't bother if it already exists.
	{
		image_width = tex->width;
		image_height = tex->height;
		return tex;
	}

	tex = R_LoadCompressed(name);
	if (TEXVALID(tex))
		return tex;

	if (strchr(name, '/'))	//never look in a root dir for the pic
		i = 0;
	else
		i = 1;

	//should write this nicer.
	for (; i < sizeof(tex_path)/sizeof(tex_path[0]); i++)
	{
		if (!tex_path[i].enabled)
			continue;
		for (e = sizeof(extensions)/sizeof(char *)-1; e >=0 ; e--)
		{
			size_t fsize;
			if (tex_path[i].args >= 3)
			{
				if (!subpath)
					continue;
				snprintf(fname, sizeof(fname)-1, tex_path[i].path, subpath, nicename, extensions[e]);
			}
			else
				snprintf(fname, sizeof(fname)-1, tex_path[i].path, nicename, extensions[e]);

			TRACE(("dbg: Mod_LoadBumpmapTexture: opening %s\n", fname));

			if ((buf = FS_LoadMallocFile (fname, &fsize)))
			{
				if ((data = ReadTargaFile(buf, fsize, &image_width, &image_height, &hasalpha, 2)))	//Only load a greyscale image.
				{
					TRACE(("dbg: Mod_LoadBumpmapTexture: tga %s loaded\n", name));
					TEXASSIGNF(tex, R_LoadTexture8Bump(name, image_width, image_height, data, IF_NOALPHA|IF_NOGAMMA));
					BZ_Free(data);
				}
				else
				{
					BZ_Free(buf);
					continue;
				}

				BZ_Free(buf);

				return tex;
			}
		}
	}
	return r_nulltex;
}
#endif

// ocrana led functions
static int ledcolors[8][3] =
{
	// green
	{ 0, 255, 0 },
	{ 0, 127, 0 },
	// red
	{ 255, 0, 0 },
	{ 127, 0, 0 },
	// yellow
	{ 255, 255, 0 },
	{ 127, 127, 0 },
	// blue
	{ 0, 0, 255 },
	{ 0, 0, 127 }
};

void AddOcranaLEDsIndexed (qbyte *image, int h, int w)
{
	int tridx[8]; // transition indexes
	qbyte *point;
	int i, idx, x, y, rs;
	int r, g, b, rd, gd, bd;

	// calc row size, character size
	rs = w;
	h /= 16;
	w /= 16;

	// generate palettes
	for (i = 0; i < 4; i++)
	{
		// get palette
		r = ledcolors[i*2][0];
		g = ledcolors[i*2][1];
		b = ledcolors[i*2][2];
		rd = (r - ledcolors[i*2+1][0]) / 8;
		gd = (g - ledcolors[i*2+1][1]) / 8;
		bd = (b - ledcolors[i*2+1][2]) / 8;
		for (idx = 0; idx < 8; idx++)
		{
			tridx[idx] = GetPaletteIndex(r, g, b);
			r -= rd;
			g -= gd;
			b -= bd;
		}

		// generate LED into image
		b = (w * w + h * h) / 16;
		if (b < 1)
			b = 1;
		rd = w + 1;
		gd = h + 1;

		point = image + (8 * rs * h) + ((6 + i) * w);
		for (y = 1; y <= h; y++)
		{
			for (x = 1; x <= w; x++)
			{
				r = rd - (x*2); r *= r;
				g = gd - (y*2); g *= g;
				idx = (r + g) / b;

				if (idx > 7)
					*point++ = 0;
				else
					*point++ = tridx[idx];
			}
			point += rs - w;
		}
	}
}
#endif

/*
Find closest color in the palette for named color
*/
int MipColor(int r, int g, int b)
{
	int i;
	float dist;
	int best=15;
	float bestdist;
	int r1, g1, b1;
	static int lr = -1, lg = -1, lb = -1;
	static int lastbest;

	if (r == lr && g == lg && b == lb)
		return lastbest;

	bestdist = 256*256*3;

	for (i = 0; i < 256; i++)
	{
		r1 = host_basepal[i*3] - r;
		g1 = host_basepal[i*3+1] - g;
		b1 = host_basepal[i*3+2] - b;
		dist = r1*r1 + g1*g1 + b1*b1;
		if (dist < bestdist) {
			bestdist = dist;
			best = i;
		}
	}
	lr = r; lg = g; lb = b;
	lastbest = best;
	return best;
}

qboolean SCR_ScreenShot (char *filename, enum fs_relative fsroot, void **buffer, int numbuffers, qintptr_t bytestride, int width, int height, enum uploadfmt fmt, qboolean writemeta)
{
	char ext[8];
	void *nbuffers[2];

	switch(fmt)
	{	//nuke any alpha channel...
	case TF_RGBA32: fmt = TF_RGBX32; break;
	case TF_BGRA32: fmt = TF_BGRX32; break;
	default: break;
	}

	if (!bytestride)
		bytestride = width*4;
	if (bytestride < 0)
	{	//fix up the buffers so callers don't have to.
		int nb = numbuffers;
		for (numbuffers = 0; numbuffers < nb && numbuffers < countof(nbuffers); numbuffers++)
		nbuffers[numbuffers] = (char*)buffer[numbuffers] - bytestride*(height-1);
		buffer = nbuffers;
	}

	COM_FileExtension(filename, ext, sizeof(ext));

#ifdef AVAIL_PNGLIB
	if (!Q_strcasecmp(ext, "png") || !Q_strcasecmp(ext, "pns"))
	{
		//png can do bgr+rgb
		//rgba bgra will result in an extra alpha chan
		//actual stereo is also supported. huzzah.
		return Image_WritePNG(filename, fsroot, scr_sshot_compression.value, buffer, numbuffers, bytestride, width, height, fmt, writemeta);
	}
	else
#endif
#ifdef AVAIL_JPEGLIB
	if (!Q_strcasecmp(ext, "jpeg") || !Q_strcasecmp(ext, "jpg") || !Q_strcasecmp(ext, "jps"))
	{
		return screenshotJPEG(filename, fsroot, scr_sshot_compression.value, buffer[0], bytestride, width, height, fmt, writemeta);
	}
	else
#endif
#ifdef IMAGEFMT_BMP
	if (!Q_strcasecmp(ext, "bmp"))
	{
		return WriteBMPFile(filename, fsroot, buffer[0], bytestride, width, height, fmt);
	}
	else
#endif
#ifdef IMAGEFMT_PCX
	if (!Q_strcasecmp(ext, "pcx"))
	{
		int y, x, s;
		qbyte *src, *dest;
		qbyte *srcbuf = buffer[0], *dstbuf;
		if (fmt == TF_RGB24 || fmt == TF_RGBA32 || fmt == TF_RGBX32 || fmt == PTI_LLLA8 || fmt == PTI_LLLX8)
		{
			dstbuf = malloc((intptr_t)width*height);
			s = (fmt == TF_RGB24)?3:4;
			// convert in-place to eight bit
			for (y = 0; y < height; y++)
			{
				src = srcbuf + (bytestride * y);
				dest = dstbuf + (width * y);

				for (x = 0; x < width; x++) {
					*dest++ = MipColor(src[0], src[1], src[2]);
					src += s;
				}
			}
		}
		else if (fmt == TF_BGR24 || fmt == TF_BGRA32 || fmt == TF_BGRX32)
		{
			dstbuf = malloc((intptr_t)width*height);
			s = (fmt == TF_BGR24)?3:4;
			// convert in-place to eight bit
			for (y = 0; y < height; y++)
			{
				src = srcbuf + (bytestride * y);
				dest = dstbuf + (width * y);

				for (x = 0; x < width; x++) {
					*dest++ = MipColor(src[2], src[1], src[0]);
					src += s;
				}
			}
		}
		else
			return false;

		WritePCXfile (filename, fsroot, dstbuf, width, height, width, host_basepal, false);
		free(dstbuf);
	}
	else
#endif
	if (!Q_strcasecmp(ext, "tga"))	//tga
		return WriteTGA(filename, fsroot, buffer[0], bytestride, width, height, fmt);
#ifdef IMAGEFMT_KTX
	else if (!Q_strcasecmp(ext, "ktx") && bytestride > 0)	//ktx
	{
		struct pendingtextureinfo out = {PTI_2D};
		out.encoding = fmt;
		out.mipcount = 1;
		out.mip[0].data = buffer[0];
		out.mip[0].datasize = bytestride*height;
		out.mip[0].width = width;
		out.mip[0].height = height;
		out.mip[0].depth = 1;
		return Image_WriteKTXFile(filename, fsroot, &out);
	}
#endif
#ifdef IMAGEFMT_DDS
	else if (!Q_strcasecmp(ext, "dds") && bytestride > 0)	//dds
	{
		struct pendingtextureinfo out = {PTI_2D};
		out.encoding = fmt;
		out.mipcount = 1;
		out.mip[0].data = buffer[0];
		out.mip[0].datasize = bytestride*height;
		out.mip[0].width = width;
		out.mip[0].height = height;
		out.mip[0].depth = 1;
		return Image_WriteDDSFile(filename, fsroot, &out);
	}
#endif
	else	//extension / type not recognised.
		return false;
	return true;
}