/* $Id: tif_luv.c,v 1.17.2.4 2010-06-08 18:50:42 bfriesen Exp $ */

/*

 * Copyright (c) 1997 Greg Ward Larson

 * Copyright (c) 1997 Silicon Graphics, Inc.

 *

 * Permission to use, copy, modify, distribute, and sell this software and 

 * its documentation for any purpose is hereby granted without fee, provided

 * that (i) the above copyright notices and this permission notice appear in

 * all copies of the software and related documentation, and (ii) the names of

 * Sam Leffler, Greg Larson and Silicon Graphics may not be used in any

 * advertising or publicity relating to the software without the specific,

 * prior written permission of Sam Leffler, Greg Larson and Silicon Graphics.

 * 

 * THE SOFTWARE IS PROVIDED "AS-IS" AND WITHOUT WARRANTY OF ANY KIND, 

 * EXPRESS, IMPLIED OR OTHERWISE, INCLUDING WITHOUT LIMITATION, ANY 

 * WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.  

 * 

 * IN NO EVENT SHALL SAM LEFFLER, GREG LARSON OR SILICON GRAPHICS BE LIABLE

 * FOR ANY SPECIAL, INCIDENTAL, INDIRECT OR CONSEQUENTIAL DAMAGES OF ANY KIND,

 * OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS,

 * WHETHER OR NOT ADVISED OF THE POSSIBILITY OF DAMAGE, AND ON ANY THEORY OF 

 * LIABILITY, ARISING OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE 

 * OF THIS SOFTWARE.

 */

#include "tiffiop.h"

#ifdef LOGLUV_SUPPORT

/*

 * TIFF Library.

 * LogLuv compression support for high dynamic range images.

 *

 * Contributed by Greg Larson.

 *

 * LogLuv image support uses the TIFF library to store 16 or 10-bit

 * log luminance values with 8 bits each of u and v or a 14-bit index.

 *

 * The codec can take as input and produce as output 32-bit IEEE float values 

 * as well as 16-bit integer values.  A 16-bit luminance is interpreted

 * as a sign bit followed by a 15-bit integer that is converted

 * to and from a linear magnitude using the transformation:

 *

 *	L = 2^( (Le+.5)/256 - 64 )		# real from 15-bit

 *

 *	Le = floor( 256*(log2(L) + 64) )	# 15-bit from real

 *

 * The actual conversion to world luminance units in candelas per sq. meter

 * requires an additional multiplier, which is stored in the TIFFTAG_STONITS.

 * This value is usually set such that a reasonable exposure comes from

 * clamping decoded luminances above 1 to 1 in the displayed image.

 *

 * The 16-bit values for u and v may be converted to real values by dividing

 * each by 32768.  (This allows for negative values, which aren't useful as

 * far as we know, but are left in case of future improvements in human

 * color vision.)

 *

 * Conversion from (u,v), which is actually the CIE (u',v') system for

 * you color scientists, is accomplished by the following transformation:

 *

 *	u = 4*x / (-2*x + 12*y + 3)

 *	v = 9*y / (-2*x + 12*y + 3)

 *

 *	x = 9*u / (6*u - 16*v + 12)

 *	y = 4*v / (6*u - 16*v + 12)

 *

 * This process is greatly simplified by passing 32-bit IEEE floats

 * for each of three CIE XYZ coordinates.  The codec then takes care

 * of conversion to and from LogLuv, though the application is still

 * responsible for interpreting the TIFFTAG_STONITS calibration factor.

 *

 * By definition, a CIE XYZ vector of [1 1 1] corresponds to a neutral white

 * point of (x,y)=(1/3,1/3).  However, most color systems assume some other

 * white point, such as D65, and an absolute color conversion to XYZ then

 * to another color space with a different white point may introduce an

 * unwanted color cast to the image.  It is often desirable, therefore, to

 * perform a white point conversion that maps the input white to [1 1 1]

 * in XYZ, then record the original white point using the TIFFTAG_WHITEPOINT

 * tag value.  A decoder that demands absolute color calibration may use

 * this white point tag to get back the original colors, but usually it

 * will be ignored and the new white point will be used instead that

 * matches the output color space.

 *

 * Pixel information is compressed into one of two basic encodings, depending

 * on the setting of the compression tag, which is one of COMPRESSION_SGILOG

 * or COMPRESSION_SGILOG24.  For COMPRESSION_SGILOG, greyscale data is

 * stored as:

 *

 *	 1       15

 *	|-+---------------|

 *

 * COMPRESSION_SGILOG color data is stored as:

 *

 *	 1       15           8        8

 *	|-+---------------|--------+--------|

 *	 S       Le           ue       ve

 *

 * For the 24-bit COMPRESSION_SGILOG24 color format, the data is stored as:

 *

 *	     10           14

 *	|----------|--------------|

 *	     Le'          Ce

 *

 * There is no sign bit in the 24-bit case, and the (u,v) chromaticity is

 * encoded as an index for optimal color resolution.  The 10 log bits are

 * defined by the following conversions:

 *

 *	L = 2^((Le'+.5)/64 - 12)		# real from 10-bit

 *

 *	Le' = floor( 64*(log2(L) + 12) )	# 10-bit from real

 *

 * The 10 bits of the smaller format may be converted into the 15 bits of

 * the larger format by multiplying by 4 and adding 13314.  Obviously,

 * a smaller range of magnitudes is covered (about 5 orders of magnitude

 * instead of 38), and the lack of a sign bit means that negative luminances

 * are not allowed.  (Well, they aren't allowed in the real world, either,

 * but they are useful for certain types of image processing.)

 *

 * The desired user format is controlled by the setting the internal

 * pseudo tag TIFFTAG_SGILOGDATAFMT to one of:

 *  SGILOGDATAFMT_FLOAT       = IEEE 32-bit float XYZ values

 *  SGILOGDATAFMT_16BIT	      = 16-bit integer encodings of logL, u and v

 * Raw data i/o is also possible using:

 *  SGILOGDATAFMT_RAW         = 32-bit unsigned integer with encoded pixel

 * In addition, the following decoding is provided for ease of display:

 *  SGILOGDATAFMT_8BIT        = 8-bit default RGB gamma-corrected values

 *

 * For grayscale images, we provide the following data formats:

 *  SGILOGDATAFMT_FLOAT       = IEEE 32-bit float Y values

 *  SGILOGDATAFMT_16BIT       = 16-bit integer w/ encoded luminance

 *  SGILOGDATAFMT_8BIT        = 8-bit gray monitor values

 *

 * Note that the COMPRESSION_SGILOG applies a simple run-length encoding

 * scheme by separating the logL, u and v bytes for each row and applying

 * a PackBits type of compression.  Since the 24-bit encoding is not

 * adaptive, the 32-bit color format takes less space in many cases.

 *

 * Further control is provided over the conversion from higher-resolution

 * formats to final encoded values through the pseudo tag

 * TIFFTAG_SGILOGENCODE:

 *  SGILOGENCODE_NODITHER     = do not dither encoded values

 *  SGILOGENCODE_RANDITHER    = apply random dithering during encoding

 *

 * The default value of this tag is SGILOGENCODE_NODITHER for

 * COMPRESSION_SGILOG to maximize run-length encoding and

 * SGILOGENCODE_RANDITHER for COMPRESSION_SGILOG24 to turn

 * quantization errors into noise.

 */

#include <stdio.h>

#include <stdlib.h>

#include <math.h>

/*

 * State block for each open TIFF

 * file using LogLuv compression/decompression.

 */

typedef	struct logLuvState LogLuvState;

struct logLuvState {

	int			user_datafmt;	/* user data format */

	int			encode_meth;	/* encoding method */

	int			pixel_size;	/* bytes per pixel */

	tidata_t*		tbuf;		/* translation buffer */

	int			tbuflen;	/* buffer length */

	void (*tfunc)(LogLuvState*, tidata_t, int);

	TIFFVSetMethod		vgetparent;	/* super-class method */

	TIFFVSetMethod		vsetparent;	/* super-class method */

};

#define	DecoderState(tif)	((LogLuvState*) (tif)->tif_data)

#define	EncoderState(tif)	((LogLuvState*) (tif)->tif_data)

#define SGILOGDATAFMT_UNKNOWN	-1

#define MINRUN		4	/* minimum run length */

/*

 * Decode a string of 16-bit gray pixels.

 */

static int

LogL16Decode(TIFF* tif, tidata_t op, tsize_t occ, tsample_t s)

{

	LogLuvState* sp = DecoderState(tif);

	int shft, i, npixels;

	unsigned char* bp;

	int16* tp;

	int16 b;

	int cc, rc;

	assert(s == 0);

	assert(sp != NULL);

	npixels = occ / sp->pixel_size;

	if (sp->user_datafmt == SGILOGDATAFMT_16BIT)

		tp = (int16*) op;

	else {

		assert(sp->tbuflen >= npixels);

		tp = (int16*) sp->tbuf;

	}

	_TIFFmemset((tdata_t) tp, 0, npixels*sizeof (tp[0]));

	bp = (unsigned char*) tif->tif_rawcp;

	cc = tif->tif_rawcc;

					/* get each byte string */

	for (shft = 2*8; (shft -= 8) >= 0; ) {

		for (i = 0; i < npixels && cc > 0; )

			if (*bp >= 128) {		/* run */

				rc = *bp++ + (2-128);

				b = (int16)(*bp++ << shft);

				cc -= 2;

				while (rc-- && i < npixels)

					tp[i++] |= b;

			} else {			/* non-run */

				rc = *bp++;		/* nul is noop */

				while (--cc && rc-- && i < npixels)

					tp[i++] |= (int16)*bp++ << shft;

			}

		if (i != npixels) {

			TIFFErrorExt(tif->tif_clientdata, tif->tif_name,

		"LogL16Decode: Not enough data at row %d (short %d pixels)",

			    tif->tif_row, npixels - i);

			tif->tif_rawcp = (tidata_t) bp;

			tif->tif_rawcc = cc;

			return (0);

		}

	}

	(*sp->tfunc)(sp, op, npixels);

	tif->tif_rawcp = (tidata_t) bp;

	tif->tif_rawcc = cc;

	return (1);

}

/*

 * Decode a string of 24-bit pixels.

 */

static int

LogLuvDecode24(TIFF* tif, tidata_t op, tsize_t occ, tsample_t s)

{

	LogLuvState* sp = DecoderState(tif);

	int cc, i, npixels;

	unsigned char* bp;

	uint32* tp;

	assert(s == 0);

	assert(sp != NULL);

	npixels = occ / sp->pixel_size;

	if (sp->user_datafmt == SGILOGDATAFMT_RAW)

		tp = (uint32 *)op;

	else {

		assert(sp->tbuflen >= npixels);

		tp = (uint32 *) sp->tbuf;

	}

					/* copy to array of uint32 */

	bp = (unsigned char*) tif->tif_rawcp;

	cc = tif->tif_rawcc;

	for (i = 0; i < npixels && cc > 0; i++) {

		tp[i] = bp[0] << 16 | bp[1] << 8 | bp[2];

		bp += 3;

		cc -= 3;

	}

	tif->tif_rawcp = (tidata_t) bp;

	tif->tif_rawcc = cc;

	if (i != npixels) {

		TIFFErrorExt(tif->tif_clientdata, tif->tif_name,

	    "LogLuvDecode24: Not enough data at row %d (short %d pixels)",

		    tif->tif_row, npixels - i);

		return (0);

	}

	(*sp->tfunc)(sp, op, npixels);

	return (1);

}

/*

 * Decode a string of 32-bit pixels.

 */

static int

LogLuvDecode32(TIFF* tif, tidata_t op, tsize_t occ, tsample_t s)

{

	LogLuvState* sp;

	int shft, i, npixels;

	unsigned char* bp;

	uint32* tp;

	uint32 b;

	int cc, rc;

	assert(s == 0);

	sp = DecoderState(tif);

	assert(sp != NULL);

	npixels = occ / sp->pixel_size;

	if (sp->user_datafmt == SGILOGDATAFMT_RAW)

		tp = (uint32*) op;

	else {

		assert(sp->tbuflen >= npixels);

		tp = (uint32*) sp->tbuf;

	}

	_TIFFmemset((tdata_t) tp, 0, npixels*sizeof (tp[0]));

	bp = (unsigned char*) tif->tif_rawcp;

	cc = tif->tif_rawcc;

					/* get each byte string */

	for (shft = 4*8; (shft -= 8) >= 0; ) {

		for (i = 0; i < npixels && cc > 0; )

			if (*bp >= 128) {		/* run */

				rc = *bp++ + (2-128);

				b = (uint32)*bp++ << shft;

				cc -= 2;

				while (rc-- && i < npixels)

					tp[i++] |= b;

			} else {			/* non-run */

				rc = *bp++;		/* nul is noop */

				while (--cc && rc-- && i < npixels)

					tp[i++] |= (uint32)*bp++ << shft;

			}

		if (i != npixels) {

			TIFFErrorExt(tif->tif_clientdata, tif->tif_name,

		"LogLuvDecode32: Not enough data at row %d (short %d pixels)",

			    tif->tif_row, npixels - i);

			tif->tif_rawcp = (tidata_t) bp;

			tif->tif_rawcc = cc;

			return (0);

		}

	}

	(*sp->tfunc)(sp, op, npixels);

	tif->tif_rawcp = (tidata_t) bp;

	tif->tif_rawcc = cc;

	return (1);

}

/*

 * Decode a strip of pixels.  We break it into rows to

 * maintain synchrony with the encode algorithm, which

 * is row by row.

 */

static int

LogLuvDecodeStrip(TIFF* tif, tidata_t bp, tsize_t cc, tsample_t s)

{

	tsize_t rowlen = TIFFScanlineSize(tif);

	assert(cc%rowlen == 0);

	while (cc && (*tif->tif_decoderow)(tif, bp, rowlen, s))

		bp += rowlen, cc -= rowlen;

	return (cc == 0);

}

/*

 * Decode a tile of pixels.  We break it into rows to

 * maintain synchrony with the encode algorithm, which

 * is row by row.

 */

static int

LogLuvDecodeTile(TIFF* tif, tidata_t bp, tsize_t cc, tsample_t s)

{

	tsize_t rowlen = TIFFTileRowSize(tif);

	assert(cc%rowlen == 0);

	while (cc && (*tif->tif_decoderow)(tif, bp, rowlen, s))

		bp += rowlen, cc -= rowlen;

	return (cc == 0);

}

/*

 * Encode a row of 16-bit pixels.

 */

static int

LogL16Encode(TIFF* tif, tidata_t bp, tsize_t cc, tsample_t s)

{

	LogLuvState* sp = EncoderState(tif);

	int shft, i, j, npixels;

	tidata_t op;

	int16* tp;

	int16 b;

	int occ, rc=0, mask, beg;

	assert(s == 0);

	assert(sp != NULL);

	npixels = cc / sp->pixel_size;

	if (sp->user_datafmt == SGILOGDATAFMT_16BIT)

		tp = (int16*) bp;

	else {

		tp = (int16*) sp->tbuf;

		assert(sp->tbuflen >= npixels);

		(*sp->tfunc)(sp, bp, npixels);

	}

					/* compress each byte string */

	op = tif->tif_rawcp;

	occ = tif->tif_rawdatasize - tif->tif_rawcc;

	for (shft = 2*8; (shft -= 8) >= 0; )

		for (i = 0; i < npixels; i += rc) {

			if (occ < 4) {

				tif->tif_rawcp = op;

				tif->tif_rawcc = tif->tif_rawdatasize - occ;

				if (!TIFFFlushData1(tif))

					return (-1);

				op = tif->tif_rawcp;

				occ = tif->tif_rawdatasize - tif->tif_rawcc;

			}

			mask = 0xff << shft;		/* find next run */

			for (beg = i; beg < npixels; beg += rc) {

				b = (int16) (tp[beg] & mask);

				rc = 1;

				while (rc < 127+2 && beg+rc < npixels &&

						(tp[beg+rc] & mask) == b)

					rc++;

				if (rc >= MINRUN)

					break;		/* long enough */

			}

			if (beg-i > 1 && beg-i < MINRUN) {

				b = (int16) (tp[i] & mask);/*check short run */

				j = i+1;

				while ((tp[j++] & mask) == b)

                                    if (j == beg) {

                                        *op++ = (tidataval_t)(128-2+j-i);

                                        *op++ = (tidataval_t) (b >> shft);

                                        occ -= 2;

                                        i = beg;

                                        break;

                                    }

			}

			while (i < beg) {		/* write out non-run */

				if ((j = beg-i) > 127) j = 127;

				if (occ < j+3) {

                                    tif->tif_rawcp = op;

                                    tif->tif_rawcc = tif->tif_rawdatasize - occ;

                                    if (!TIFFFlushData1(tif))

                                        return (-1);

                                    op = tif->tif_rawcp;

                                    occ = tif->tif_rawdatasize - tif->tif_rawcc;

				}

				*op++ = (tidataval_t) j; occ--;

				while (j--) {

					*op++ = (tidataval_t) (tp[i++] >> shft & 0xff);

					occ--;

				}

			}

			if (rc >= MINRUN) {		/* write out run */

				*op++ = (tidataval_t) (128-2+rc);

				*op++ = (tidataval_t) (tp[beg] >> shft & 0xff);

				occ -= 2;

			} else

				rc = 0;

		}

	tif->tif_rawcp = op;

	tif->tif_rawcc = tif->tif_rawdatasize - occ;

	return (1);

}

/*

 * Encode a row of 24-bit pixels.

 */

static int

LogLuvEncode24(TIFF* tif, tidata_t bp, tsize_t cc, tsample_t s)

{

	LogLuvState* sp = EncoderState(tif);

	int i, npixels, occ;

	tidata_t op;

	uint32* tp;

	assert(s == 0);

	assert(sp != NULL);

	npixels = cc / sp->pixel_size;

	if (sp->user_datafmt == SGILOGDATAFMT_RAW)

		tp = (uint32*) bp;

	else {

		tp = (uint32*) sp->tbuf;

		assert(sp->tbuflen >= npixels);

		(*sp->tfunc)(sp, bp, npixels);

	}

					/* write out encoded pixels */

	op = tif->tif_rawcp;

	occ = tif->tif_rawdatasize - tif->tif_rawcc;

	for (i = npixels; i--; ) {

		if (occ < 3) {

			tif->tif_rawcp = op;

			tif->tif_rawcc = tif->tif_rawdatasize - occ;

			if (!TIFFFlushData1(tif))

				return (-1);

			op = tif->tif_rawcp;

			occ = tif->tif_rawdatasize - tif->tif_rawcc;

		}

		*op++ = (tidataval_t)(*tp >> 16);

		*op++ = (tidataval_t)(*tp >> 8 & 0xff);

		*op++ = (tidataval_t)(*tp++ & 0xff);

		occ -= 3;

	}

	tif->tif_rawcp = op;

	tif->tif_rawcc = tif->tif_rawdatasize - occ;

	return (1);

}

/*

 * Encode a row of 32-bit pixels.

 */

static int

LogLuvEncode32(TIFF* tif, tidata_t bp, tsize_t cc, tsample_t s)

{

	LogLuvState* sp = EncoderState(tif);

	int shft, i, j, npixels;

	tidata_t op;

	uint32* tp;

	uint32 b;

	int occ, rc=0, mask, beg;

	assert(s == 0);

	assert(sp != NULL);

	npixels = cc / sp->pixel_size;

	if (sp->user_datafmt == SGILOGDATAFMT_RAW)

		tp = (uint32*) bp;

	else {

		tp = (uint32*) sp->tbuf;

		assert(sp->tbuflen >= npixels);

		(*sp->tfunc)(sp, bp, npixels);

	}

					/* compress each byte string */

	op = tif->tif_rawcp;

	occ = tif->tif_rawdatasize - tif->tif_rawcc;

	for (shft = 4*8; (shft -= 8) >= 0; )

		for (i = 0; i < npixels; i += rc) {

			if (occ < 4) {

				tif->tif_rawcp = op;

				tif->tif_rawcc = tif->tif_rawdatasize - occ;

				if (!TIFFFlushData1(tif))

					return (-1);

				op = tif->tif_rawcp;

				occ = tif->tif_rawdatasize - tif->tif_rawcc;

			}

			mask = 0xff << shft;		/* find next run */

			for (beg = i; beg < npixels; beg += rc) {

				b = tp[beg] & mask;

				rc = 1;

				while (rc < 127+2 && beg+rc < npixels &&

						(tp[beg+rc] & mask) == b)

					rc++;

				if (rc >= MINRUN)

					break;		/* long enough */

			}

			if (beg-i > 1 && beg-i < MINRUN) {

				b = tp[i] & mask;	/* check short run */

				j = i+1;

				while ((tp[j++] & mask) == b)

					if (j == beg) {

						*op++ = (tidataval_t)(128-2+j-i);

						*op++ = (tidataval_t)(b >> shft);

						occ -= 2;

						i = beg;

						break;

					}

			}

			while (i < beg) {		/* write out non-run */

				if ((j = beg-i) > 127) j = 127;

				if (occ < j+3) {

					tif->tif_rawcp = op;

					tif->tif_rawcc = tif->tif_rawdatasize - occ;

					if (!TIFFFlushData1(tif))

						return (-1);

					op = tif->tif_rawcp;

					occ = tif->tif_rawdatasize - tif->tif_rawcc;

				}

				*op++ = (tidataval_t) j; occ--;

				while (j--) {

					*op++ = (tidataval_t)(tp[i++] >> shft & 0xff);

					occ--;

				}

			}

			if (rc >= MINRUN) {		/* write out run */

				*op++ = (tidataval_t) (128-2+rc);

				*op++ = (tidataval_t)(tp[beg] >> shft & 0xff);

				occ -= 2;

			} else

				rc = 0;

		}

	tif->tif_rawcp = op;

	tif->tif_rawcc = tif->tif_rawdatasize - occ;

	return (1);

}

/*

 * Encode a strip of pixels.  We break it into rows to

 * avoid encoding runs across row boundaries.

 */

static int

LogLuvEncodeStrip(TIFF* tif, tidata_t bp, tsize_t cc, tsample_t s)

{

	tsize_t rowlen = TIFFScanlineSize(tif);

	assert(cc%rowlen == 0);

	while (cc && (*tif->tif_encoderow)(tif, bp, rowlen, s) == 1)

		bp += rowlen, cc -= rowlen;

	return (cc == 0);

}

/*

 * Encode a tile of pixels.  We break it into rows to

 * avoid encoding runs across row boundaries.

 */

static int

LogLuvEncodeTile(TIFF* tif, tidata_t bp, tsize_t cc, tsample_t s)

{

	tsize_t rowlen = TIFFTileRowSize(tif);

	assert(cc%rowlen == 0);

	while (cc && (*tif->tif_encoderow)(tif, bp, rowlen, s) == 1)

		bp += rowlen, cc -= rowlen;

	return (cc == 0);

}

/*

 * Encode/Decode functions for converting to and from user formats.

 */

#include "uvcode.h"

#ifndef UVSCALE

#define U_NEU		0.210526316

#define V_NEU		0.473684211

#define UVSCALE		410.

#endif

#ifndef	M_LN2

#define M_LN2		0.69314718055994530942

#endif

#ifndef M_PI

#define M_PI		3.14159265358979323846

#endif

#define log2(x)		((1./M_LN2)*log(x))

#define exp2(x)		exp(M_LN2*(x))

#define itrunc(x,m)	((m)==SGILOGENCODE_NODITHER ? \

				(int)(x) : \

				(int)((x) + rand()*(1./RAND_MAX) - .5))

#if !LOGLUV_PUBLIC

static

#endif

double

LogL16toY(int p16)		/* compute luminance from 16-bit LogL */

{

	int	Le = p16 & 0x7fff;

	double	Y;

	if (!Le)

		return (0.);

	Y = exp(M_LN2/256.*(Le+.5) - M_LN2*64.);

	return (!(p16 & 0x8000) ? Y : -Y);

}

#if !LOGLUV_PUBLIC

static

#endif

int

LogL16fromY(double Y, int em)	/* get 16-bit LogL from Y */

{

	if (Y >= 1.8371976e19)

		return (0x7fff);

	if (Y <= -1.8371976e19)

		return (0xffff);

	if (Y > 5.4136769e-20)

		return itrunc(256.*(log2(Y) + 64.), em);

	if (Y < -5.4136769e-20)

		return (~0x7fff | itrunc(256.*(log2(-Y) + 64.), em));

	return (0);

}

static void

L16toY(LogLuvState* sp, tidata_t op, int n)

{

	int16* l16 = (int16*) sp->tbuf;

	float* yp = (float*) op;

	while (n-- > 0)

		*yp++ = (float)LogL16toY(*l16++);

}

static void

L16toGry(LogLuvState* sp, tidata_t op, int n)

{

	int16* l16 = (int16*) sp->tbuf;

	uint8* gp = (uint8*) op;

	while (n-- > 0) {

		double Y = LogL16toY(*l16++);

		*gp++ = (uint8) ((Y <= 0.) ? 0 : (Y >= 1.) ? 255 : (int)(256.*sqrt(Y)));

	}

}

static void

L16fromY(LogLuvState* sp, tidata_t op, int n)

{

	int16* l16 = (int16*) sp->tbuf;

	float* yp = (float*) op;

	while (n-- > 0)

		*l16++ = (int16) (LogL16fromY(*yp++, sp->encode_meth));

}

#if !LOGLUV_PUBLIC

static

#endif

void

XYZtoRGB24(float xyz[3], uint8 rgb[3])

{

	double	r, g, b;

					/* assume CCIR-709 primaries */

	r =  2.690*xyz[0] + -1.276*xyz[1] + -0.414*xyz[2];

	g = -1.022*xyz[0] +  1.978*xyz[1] +  0.044*xyz[2];

	b =  0.061*xyz[0] + -0.224*xyz[1] +  1.163*xyz[2];

					/* assume 2.0 gamma for speed */

	/* could use integer sqrt approx., but this is probably faster */

	rgb[0] = (uint8)((r<=0.) ? 0 : (r >= 1.) ? 255 : (int)(256.*sqrt(r)));

	rgb[1] = (uint8)((g<=0.) ? 0 : (g >= 1.) ? 255 : (int)(256.*sqrt(g)));

	rgb[2] = (uint8)((b<=0.) ? 0 : (b >= 1.) ? 255 : (int)(256.*sqrt(b)));

}

#if !LOGLUV_PUBLIC

static

#endif

double

LogL10toY(int p10)		/* compute luminance from 10-bit LogL */

{

	if (p10 == 0)

		return (0.);

	return (exp(M_LN2/64.*(p10+.5) - M_LN2*12.));

}

#if !LOGLUV_PUBLIC

static

#endif

int

LogL10fromY(double Y, int em)	/* get 10-bit LogL from Y */

{

	if (Y >= 15.742)

		return (0x3ff);

	else if (Y <= .00024283)

		return (0);

	else

		return itrunc(64.*(log2(Y) + 12.), em);

}

#define NANGLES		100

#define uv2ang(u, v)	( (NANGLES*.499999999/M_PI) \

				* atan2((v)-V_NEU,(u)-U_NEU) + .5*NANGLES )

static int

oog_encode(double u, double v)		/* encode out-of-gamut chroma */

{

	static int	oog_table[NANGLES];

	static int	initialized = 0;

	register int	i;

	if (!initialized) {		/* set up perimeter table */

		double	eps[NANGLES], ua, va, ang, epsa;

		int	ui, vi, ustep;

		for (i = NANGLES; i--; )

			eps[i] = 2.;

		for (vi = UV_NVS; vi--; ) {

			va = UV_VSTART + (vi+.5)*UV_SQSIZ;

			ustep = uv_row[vi].nus-1;

			if (vi == UV_NVS-1 || vi == 0 || ustep <= 0)

				ustep = 1;

			for (ui = uv_row[vi].nus-1; ui >= 0; ui -= ustep) {

				ua = uv_row[vi].ustart + (ui+.5)*UV_SQSIZ;

				ang = uv2ang(ua, va);

                                i = (int) ang;

				epsa = fabs(ang - (i+.5));

				if (epsa < eps[i]) {

					oog_table[i] = uv_row[vi].ncum + ui;

					eps[i] = epsa;

				}

			}

		}

		for (i = NANGLES; i--; )	/* fill any holes */

			if (eps[i] > 1.5) {

				int	i1, i2;

				for (i1 = 1; i1 < NANGLES/2; i1++)

					if (eps[(i+i1)%NANGLES] < 1.5)

						break;

				for (i2 = 1; i2 < NANGLES/2; i2++)

					if (eps[(i+NANGLES-i2)%NANGLES] < 1.5)

						break;

				if (i1 < i2)

					oog_table[i] =

						oog_table[(i+i1)%NANGLES];

				else

					oog_table[i] =

						oog_table[(i+NANGLES-i2)%NANGLES];

			}

		initialized = 1;

	}

	i = (int) uv2ang(u, v);		/* look up hue angle */

	return (oog_table[i]);

}

#undef uv2ang

#undef NANGLES

#if !LOGLUV_PUBLIC

static

#endif

int

uv_encode(double u, double v, int em)	/* encode (u',v') coordinates */

{

	register int	vi, ui;

	if (v < UV_VSTART)

		return oog_encode(u, v);

	vi = itrunc((v - UV_VSTART)*(1./UV_SQSIZ), em);

	if (vi >= UV_NVS)

		return oog_encode(u, v);

	if (u < uv_row[vi].ustart)

		return oog_encode(u, v);

	ui = itrunc((u - uv_row[vi].ustart)*(1./UV_SQSIZ), em);

	if (ui >= uv_row[vi].nus)

		return oog_encode(u, v);

	return (uv_row[vi].ncum + ui);

}

#if !LOGLUV_PUBLIC

static

#endif

int

uv_decode(double *up, double *vp, int c)	/* decode (u',v') index */

{

	int	upper, lower;

	register int	ui, vi;

	if (c < 0 || c >= UV_NDIVS)

		return (-1);

	lower = 0;				/* binary search */

	upper = UV_NVS;

	while (upper - lower > 1) {

		vi = (lower + upper) >> 1;

		ui = c - uv_row[vi].ncum;

		if (ui > 0)

			lower = vi;

		else if (ui < 0)

			upper = vi;

		else {

			lower = vi;

			break;

		}

	}

	vi = lower;

	ui = c - uv_row[vi].ncum;

	*up = uv_row[vi].ustart + (ui+.5)*UV_SQSIZ;

	*vp = UV_VSTART + (vi+.5)*UV_SQSIZ;

	return (0);

}

#if !LOGLUV_PUBLIC

static

#endif

void

LogLuv24toXYZ(uint32 p, float XYZ[3])

{

	int	Ce;

	double	L, u, v, s, x, y;

					/* decode luminance */

	L = LogL10toY(p>>14 & 0x3ff);

	if (L <= 0.) {

		XYZ[0] = XYZ[1] = XYZ[2] = 0.;

		return;

	}

					/* decode color */

	Ce = p & 0x3fff;

	if (uv_decode(&u, &v, Ce) < 0) {