src/tiff/tif_pixarlog.c

/*
 * Copyright (c) 1996-1997 Sam Leffler
 * Copyright (c) 1996 Pixar
 *
 * 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
 * Pixar, Sam Leffler and Silicon Graphics may not be used in any advertising or
 * publicity relating to the software without the specific, prior written
 * permission of Pixar, Sam Leffler 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 PIXAR, SAM LEFFLER 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 PIXARLOG_SUPPORT

/*
 * TIFF Library.
 * PixarLog Compression Support
 *
 * Contributed by Dan McCoy.
 *
 * PixarLog film support uses the TIFF library to store companded
 * 11 bit values into a tiff file, which are compressed using the
 * zip compressor.
 *
 * The codec can take as input and produce as output 32-bit IEEE float values
 * as well as 16-bit or 8-bit unsigned integer values.
 *
 * On writing any of the above are converted into the internal
 * 11-bit log format.   In the case of  8 and 16 bit values, the
 * input is assumed to be unsigned linear color values that represent
 * the range 0-1.  In the case of IEEE values, the 0-1 range is assumed to
 * be the normal linear color range, in addition over 1 values are
 * accepted up to a value of about 25.0 to encode "hot" hightlights and such.
 * The encoding is lossless for 8-bit values, slightly lossy for the
 * other bit depths.  The actual color precision should be better
 * than the human eye can perceive with extra room to allow for
 * error introduced by further image computation.  As with any quantized
 * color format, it is possible to perform image calculations which
 * expose the quantization error. This format should certainly be less
 * susceptable to such errors than standard 8-bit encodings, but more
 * susceptable than straight 16-bit or 32-bit encodings.
 *
 * On reading the internal format is converted to the desired output format.
 * The program can request which format it desires by setting the internal
 * pseudo tag TIFFTAG_PIXARLOGDATAFMT to one of these possible values:
 *  PIXARLOGDATAFMT_FLOAT     = provide IEEE float values.
 *  PIXARLOGDATAFMT_16BIT     = provide unsigned 16-bit integer values
 *  PIXARLOGDATAFMT_8BIT      = provide unsigned 8-bit integer values
 *
 * alternately PIXARLOGDATAFMT_8BITABGR provides unsigned 8-bit integer
 * values with the difference that if there are exactly three or four channels
 * (rgb or rgba) it swaps the channel order (bgr or abgr).
 *
 * PIXARLOGDATAFMT_11BITLOG provides the internal encoding directly
 * packed in 16-bit values.   However no tools are supplied for interpreting
 * these values.
 *
 * "hot" (over 1.0) areas written in floating point get clamped to
 * 1.0 in the integer data types.
 *
 * When the file is closed after writing, the bit depth and sample format
 * are set always to appear as if 8-bit data has been written into it.
 * That way a naive program unaware of the particulars of the encoding
 * gets the format it is most likely able to handle.
 *
 * The codec does it's own horizontal differencing step on the coded
 * values so the libraries predictor stuff should be turned off.
 * The codec also handle byte swapping the encoded values as necessary
 * since the library does not have the information necessary
 * to know the bit depth of the raw unencoded buffer.
 *
 */

/* Watcom C++ (or its make utility) doesn't like long filenames */
#ifdef __WATCOMC__
#include "tif_pred.h"
#else
#include "tif_predict.h"
#endif

#include "zlib.h"
#include "zutil.h"

#include <stdio.h>
#include <assert.h>
#include <stdlib.h>
#include <math.h>

/* Tables for converting to/from 11 bit coded values */

#define  TSIZE	 2048		/* decode table size (11-bit tokens) */
#define  TSIZEP1 2049		/* Plus one for slop */
#define  ONE	 1250		/* token value of 1.0 exactly */
#define  RATIO	 1.004		/* nominal ratio for log part */

#define CODE_MASK 0x7ff         /* 11 bits. */

static float  Fltsize;
static float  LogK1, LogK2;

#define REPEAT(n, op)   { int i; i=n; do { i--; op; } while (i>0); }

static void
horizontalAccumulateF(uint16 *wp, int n, int stride, float *op,
	float *ToLinearF)
{
    register unsigned int  cr, cg, cb, ca, mask;
    register float  t0, t1, t2, t3;

    if (n >= stride) {
	mask = CODE_MASK;
	if (stride == 3) {
	    t0 = ToLinearF[cr = wp[0]];
	    t1 = ToLinearF[cg = wp[1]];
	    t2 = ToLinearF[cb = wp[2]];
	    op[0] = t0;
	    op[1] = t1;
	    op[2] = t2;
	    n -= 3;
	    while (n > 0) {
		wp += 3;
		op += 3;
		n -= 3;
		t0 = ToLinearF[(cr += wp[0]) & mask];
		t1 = ToLinearF[(cg += wp[1]) & mask];
		t2 = ToLinearF[(cb += wp[2]) & mask];
		op[0] = t0;
		op[1] = t1;
		op[2] = t2;
	    }
	} else if (stride == 4) {
	    t0 = ToLinearF[cr = wp[0]];
	    t1 = ToLinearF[cg = wp[1]];
	    t2 = ToLinearF[cb = wp[2]];
	    t3 = ToLinearF[ca = wp[3]];
	    op[0] = t0;
	    op[1] = t1;
	    op[2] = t2;
	    op[3] = t3;
	    n -= 4;
	    while (n > 0) {
		wp += 4;
		op += 4;
		n -= 4;
		t0 = ToLinearF[(cr += wp[0]) & mask];
		t1 = ToLinearF[(cg += wp[1]) & mask];
		t2 = ToLinearF[(cb += wp[2]) & mask];
		t3 = ToLinearF[(ca += wp[3]) & mask];
		op[0] = t0;
		op[1] = t1;
		op[2] = t2;
		op[3] = t3;
	    }
	} else {
	    REPEAT(stride, *op = ToLinearF[*wp&mask]; wp++; op++)
	    n -= stride;
	    while (n > 0) {
		REPEAT(stride,
		    wp[stride] += *wp; *op = ToLinearF[*wp&mask]; wp++; op++)
		n -= stride;
	    }
	}
    }
}

static void
horizontalAccumulate12(uint16 *wp, int n, int stride, int16 *op,
	float *ToLinearF)
{
    register unsigned int  cr, cg, cb, ca, mask;
    register float  t0, t1, t2, t3;

#define SCALE12 2048.0
#define CLAMP12(t) (((t) < 3071) ? (uint16) (t) : 3071)

    if (n >= stride) {
	mask = CODE_MASK;
	if (stride == 3) {
	    t0 = ToLinearF[cr = wp[0]] * SCALE12;
	    t1 = ToLinearF[cg = wp[1]] * SCALE12;
	    t2 = ToLinearF[cb = wp[2]] * SCALE12;
	    op[0] = CLAMP12(t0);
	    op[1] = CLAMP12(t1);
	    op[2] = CLAMP12(t2);
	    n -= 3;
	    while (n > 0) {
		wp += 3;
		op += 3;
		n -= 3;
		t0 = ToLinearF[(cr += wp[0]) & mask] * SCALE12;
		t1 = ToLinearF[(cg += wp[1]) & mask] * SCALE12;
		t2 = ToLinearF[(cb += wp[2]) & mask] * SCALE12;
		op[0] = CLAMP12(t0);
		op[1] = CLAMP12(t1);
		op[2] = CLAMP12(t2);
	    }
	} else if (stride == 4) {
	    t0 = ToLinearF[cr = wp[0]] * SCALE12;
	    t1 = ToLinearF[cg = wp[1]] * SCALE12;
	    t2 = ToLinearF[cb = wp[2]] * SCALE12;
	    t3 = ToLinearF[ca = wp[3]] * SCALE12;
	    op[0] = CLAMP12(t0);
	    op[1] = CLAMP12(t1);
	    op[2] = CLAMP12(t2);
	    op[3] = CLAMP12(t3);
	    n -= 4;
	    while (n > 0) {
		wp += 4;
		op += 4;
		n -= 4;
		t0 = ToLinearF[(cr += wp[0]) & mask] * SCALE12;
		t1 = ToLinearF[(cg += wp[1]) & mask] * SCALE12;
		t2 = ToLinearF[(cb += wp[2]) & mask] * SCALE12;
		t3 = ToLinearF[(ca += wp[3]) & mask] * SCALE12;
		op[0] = CLAMP12(t0);
		op[1] = CLAMP12(t1);
		op[2] = CLAMP12(t2);
		op[3] = CLAMP12(t3);
	    }
	} else {
	    REPEAT(stride, t0 = ToLinearF[*wp&mask] * SCALE12;
                           *op = CLAMP12(t0); wp++; op++)
	    n -= stride;
	    while (n > 0) {
		REPEAT(stride,
		    wp[stride] += *wp; t0 = ToLinearF[wp[stride]&mask]*SCALE12;
		    *op = CLAMP12(t0);  wp++; op++)
		n -= stride;
	    }
	}
    }
}

static void
horizontalAccumulate16(uint16 *wp, int n, int stride, uint16 *op,
	uint16 *ToLinear16)
{
    register unsigned int  cr, cg, cb, ca, mask;

    if (n >= stride) {
	mask = CODE_MASK;
	if (stride == 3) {
	    op[0] = ToLinear16[cr = wp[0]];
	    op[1] = ToLinear16[cg = wp[1]];
	    op[2] = ToLinear16[cb = wp[2]];
	    n -= 3;
	    while (n > 0) {
		wp += 3;
		op += 3;
		n -= 3;
		op[0] = ToLinear16[(cr += wp[0]) & mask];
		op[1] = ToLinear16[(cg += wp[1]) & mask];
		op[2] = ToLinear16[(cb += wp[2]) & mask];
	    }
	} else if (stride == 4) {
	    op[0] = ToLinear16[cr = wp[0]];
	    op[1] = ToLinear16[cg = wp[1]];
	    op[2] = ToLinear16[cb = wp[2]];
	    op[3] = ToLinear16[ca = wp[3]];
	    n -= 4;
	    while (n > 0) {
		wp += 4;
		op += 4;
		n -= 4;
		op[0] = ToLinear16[(cr += wp[0]) & mask];
		op[1] = ToLinear16[(cg += wp[1]) & mask];
		op[2] = ToLinear16[(cb += wp[2]) & mask];
		op[3] = ToLinear16[(ca += wp[3]) & mask];
	    }
	} else {
	    REPEAT(stride, *op = ToLinear16[*wp&mask]; wp++; op++)
	    n -= stride;
	    while (n > 0) {
		REPEAT(stride,
		    wp[stride] += *wp; *op = ToLinear16[*wp&mask]; wp++; op++)
		n -= stride;
	    }
	}
    }
}

/*
 * Returns the log encoded 11-bit values with the horizontal
 * differencing undone.
 */
static void
horizontalAccumulate11(uint16 *wp, int n, int stride, uint16 *op)
{
    register unsigned int  cr, cg, cb, ca, mask;

    if (n >= stride) {
	mask = CODE_MASK;
	if (stride == 3) {
	    op[0] = cr = wp[0];  op[1] = cg = wp[1];  op[2] = cb = wp[2];
	    n -= 3;
	    while (n > 0) {
		wp += 3;
		op += 3;
		n -= 3;
		op[0] = (cr += wp[0]) & mask;
		op[1] = (cg += wp[1]) & mask;
		op[2] = (cb += wp[2]) & mask;
	    }
	} else if (stride == 4) {
	    op[0] = cr = wp[0];  op[1] = cg = wp[1];
	    op[2] = cb = wp[2];  op[3] = ca = wp[3];
	    n -= 4;
	    while (n > 0) {
		wp += 4;
		op += 4;
		n -= 4;
		op[0] = (cr += wp[0]) & mask;
		op[1] = (cg += wp[1]) & mask;
		op[2] = (cb += wp[2]) & mask;
		op[3] = (ca += wp[3]) & mask;
	    }
	} else {
	    REPEAT(stride, *op = *wp&mask; wp++; op++)
	    n -= stride;
	    while (n > 0) {
		REPEAT(stride,
		    wp[stride] += *wp; *op = *wp&mask; wp++; op++)
	    	n -= stride;
	    }
	}
    }
}

static void
horizontalAccumulate8(uint16 *wp, int n, int stride, unsigned char *op,
	unsigned char *ToLinear8)
{
    register unsigned int  cr, cg, cb, ca, mask;

    if (n >= stride) {
	mask = CODE_MASK;
	if (stride == 3) {
	    op[0] = ToLinear8[cr = wp[0]];
	    op[1] = ToLinear8[cg = wp[1]];
	    op[2] = ToLinear8[cb = wp[2]];
	    n -= 3;
	    while (n > 0) {
		n -= 3;
		wp += 3;
		op += 3;
		op[0] = ToLinear8[(cr += wp[0]) & mask];
		op[1] = ToLinear8[(cg += wp[1]) & mask];
		op[2] = ToLinear8[(cb += wp[2]) & mask];
	    }
	} else if (stride == 4) {
	    op[0] = ToLinear8[cr = wp[0]];
	    op[1] = ToLinear8[cg = wp[1]];
	    op[2] = ToLinear8[cb = wp[2]];
	    op[3] = ToLinear8[ca = wp[3]];
	    n -= 4;
	    while (n > 0) {
		n -= 4;
		wp += 4;
		op += 4;
		op[0] = ToLinear8[(cr += wp[0]) & mask];
		op[1] = ToLinear8[(cg += wp[1]) & mask];
		op[2] = ToLinear8[(cb += wp[2]) & mask];
		op[3] = ToLinear8[(ca += wp[3]) & mask];
	    }
	} else {
	    REPEAT(stride, *op = ToLinear8[*wp&mask]; wp++; op++)
	    n -= stride;
	    while (n > 0) {
		REPEAT(stride,
		    wp[stride] += *wp; *op = ToLinear8[*wp&mask]; wp++; op++)
		n -= stride;
	    }
	}
    }
}


static void
horizontalAccumulate8abgr(uint16 *wp, int n, int stride, unsigned char *op,
	unsigned char *ToLinear8)
{
    register unsigned int  cr, cg, cb, ca, mask;
    register unsigned char  t0, t1, t2, t3;

    if (n >= stride) {
	mask = CODE_MASK;
	if (stride == 3) {
	    op[0] = 0;
	    t1 = ToLinear8[cb = wp[2]];
	    t2 = ToLinear8[cg = wp[1]];
	    t3 = ToLinear8[cr = wp[0]];
	    op[1] = t1;
	    op[2] = t2;
	    op[3] = t3;
	    n -= 3;
	    while (n > 0) {
		n -= 3;
		wp += 3;
		op += 4;
		op[0] = 0;
		t1 = ToLinear8[(cb += wp[2]) & mask];
		t2 = ToLinear8[(cg += wp[1]) & mask];
		t3 = ToLinear8[(cr += wp[0]) & mask];
		op[1] = t1;
		op[2] = t2;
		op[3] = t3;
	    }
	} else if (stride == 4) {
	    t0 = ToLinear8[ca = wp[3]];
	    t1 = ToLinear8[cb = wp[2]];
	    t2 = ToLinear8[cg = wp[1]];
	    t3 = ToLinear8[cr = wp[0]];
	    op[0] = t0;
	    op[1] = t1;
	    op[2] = t2;
	    op[3] = t3;
	    n -= 4;
	    while (n > 0) {
		n -= 4;
		wp += 4;
		op += 4;
		t0 = ToLinear8[(ca += wp[3]) & mask];
		t1 = ToLinear8[(cb += wp[2]) & mask];
		t2 = ToLinear8[(cg += wp[1]) & mask];
		t3 = ToLinear8[(cr += wp[0]) & mask];
		op[0] = t0;
		op[1] = t1;
		op[2] = t2;
		op[3] = t3;
	    }
	} else {
	    REPEAT(stride, *op = ToLinear8[*wp&mask]; wp++; op++)
	    n -= stride;
	    while (n > 0) {
		REPEAT(stride,
		    wp[stride] += *wp; *op = ToLinear8[*wp&mask]; wp++; op++)
		n -= stride;
	    }
	}
    }
}

/*
 * State block for each open TIFF
 * file using PixarLog compression/decompression.
 */
typedef	struct {
	TIFFPredictorState	predict;
	z_stream		stream;
	uint16			*tbuf;
	uint16			stride;
	int			state;
	int			user_datafmt;
	int			quality;
#define PLSTATE_INIT 1

	TIFFVSetMethod		vgetparent;	/* super-class method */
	TIFFVSetMethod		vsetparent;	/* super-class method */

	float *ToLinearF;
	uint16 *ToLinear16;
	unsigned char *ToLinear8;
	uint16  *FromLT2;
	uint16  *From14; /* Really for 16-bit data, but we shift down 2 */
	uint16  *From8;

} PixarLogState;

static int
PixarLogMakeTables(PixarLogState *sp)
{

/*
 *    We make several tables here to convert between various external
 *    representations (float, 16-bit, and 8-bit) and the internal
 *    11-bit companded representation.  The 11-bit representation has two
 *    distinct regions.  A linear bottom end up through .018316 in steps
 *    of about .000073, and a region of constant ratio up to about 25.
 *    These floating point numbers are stored in the main table ToLinearF.
 *    All other tables are derived from this one.  The tables (and the
 *    ratios) are continuous at the internal seam.
 */

    int  nlin, lt2size;
    int  i, j;
    double  b, c, linstep, max;
    double  k, v, dv, r, lr2, r2;
    float *ToLinearF;
    uint16 *ToLinear16;
    unsigned char *ToLinear8;
    uint16  *FromLT2;
    uint16  *From14; /* Really for 16-bit data, but we shift down 2 */
    uint16  *From8;

    c = log(RATIO);
    nlin = 1./c;	/* nlin must be an integer */
    c = 1./nlin;
    b = exp(-c*ONE);	/* multiplicative scale factor [b*exp(c*ONE) = 1] */
    linstep = b*c*exp(1.);

    LogK1 = 1./c;	/* if (v >= 2)  token = k1*log(v*k2) */
    LogK2 = 1./b;
    lt2size = (2./linstep)+1;
    FromLT2 = (uint16 *)_TIFFmalloc(lt2size*sizeof(uint16));
    From14 = (uint16 *)_TIFFmalloc(16384*sizeof(uint16));
    From8 = (uint16 *)_TIFFmalloc(256*sizeof(uint16));
    ToLinearF = (float *)_TIFFmalloc(TSIZEP1 * sizeof(float));
    ToLinear16 = (uint16 *)_TIFFmalloc(TSIZEP1 * sizeof(uint16));
    ToLinear8 = (unsigned char *)_TIFFmalloc(TSIZEP1 * sizeof(unsigned char));
    if (FromLT2 == NULL || From14  == NULL || From8   == NULL ||
	 ToLinearF == NULL || ToLinear16 == NULL || ToLinear8 == NULL) {
	if (FromLT2) _TIFFfree(FromLT2);
	if (From14) _TIFFfree(From14);
	if (From8) _TIFFfree(From8);
	if (ToLinearF) _TIFFfree(ToLinearF);
	if (ToLinear16) _TIFFfree(ToLinear16);
	if (ToLinear8) _TIFFfree(ToLinear8);
	sp->FromLT2 = NULL;
	sp->From14 = NULL;
	sp->From8 = NULL;
	sp->ToLinearF = NULL;
	sp->ToLinear16 = NULL;
	sp->ToLinear8 = NULL;
	return 0;
    }

    j = 0;

    for (i = 0; i < nlin; i++)  {
	v = i * linstep;
	ToLinearF[j++] = v;
    }

    for (i = nlin; i < TSIZE; i++)
	ToLinearF[j++] = b*exp(c*i);

    ToLinearF[2048] = ToLinearF[2047];

    for (i = 0; i < TSIZEP1; i++)  {
	v = ToLinearF[i]*65535.0 + 0.5;
	ToLinear16[i] = (v > 65535.0) ? 65535 : v;
	v = ToLinearF[i]*255.0  + 0.5;
	ToLinear8[i]  = (v > 255.0) ? 255 : v;
    }

    j = 0;
    for (i = 0; i < lt2size; i++)  {
	if ((i*linstep)*(i*linstep) > ToLinearF[j]*ToLinearF[j+1])
	    j++;
	FromLT2[i] = j;
    }

    /*
     * Since we lose info anyway on 16-bit data, we set up a 14-bit
     * table and shift 16-bit values down two bits on input.
     * saves a little table space.
     */
    j = 0;
    for (i = 0; i < 16384; i++)  {
	while ((i/16383.)*(i/16383.) > ToLinearF[j]*ToLinearF[j+1])
	    j++;
	From14[i] = j;
    }

    j = 0;
    for (i = 0; i < 256; i++)  {
	while ((i/255.)*(i/255.) > ToLinearF[j]*ToLinearF[j+1])
	    j++;
	From8[i] = j;
    }

    Fltsize = lt2size/2;

    sp->ToLinearF = ToLinearF;
    sp->ToLinear16 = ToLinear16;
    sp->ToLinear8 = ToLinear8;
    sp->FromLT2 = FromLT2;
    sp->From14 = From14;
    sp->From8 = From8;

    return 1;
}

#define	DecoderState(tif)	((PixarLogState*) (tif)->tif_data)
#define	EncoderState(tif)	((PixarLogState*) (tif)->tif_data)

static	int PixarLogEncode(TIFF*, tidata_t, tsize_t, tsample_t);
static	int PixarLogDecode(TIFF*, tidata_t, tsize_t, tsample_t);

#define N(a)   (sizeof(a)/sizeof(a[0]))
#define PIXARLOGDATAFMT_UNKNOWN	-1

static int
PixarLogGuessDataFmt(TIFFDirectory *td)
{
	int guess = PIXARLOGDATAFMT_UNKNOWN;
	int format = td->td_sampleformat;

	/* If the user didn't tell us his datafmt,
	 * take our best guess from the bitspersample.
	 */
	switch (td->td_bitspersample) {
	 case 32:
		if (format == SAMPLEFORMAT_IEEEFP)
			guess = PIXARLOGDATAFMT_FLOAT;
		break;
	 case 16:
		if (format == SAMPLEFORMAT_VOID || format == SAMPLEFORMAT_UINT)
			guess = PIXARLOGDATAFMT_16BIT;
		break;
	 case 12:
		if (format == SAMPLEFORMAT_VOID || format == SAMPLEFORMAT_INT)
			guess = PIXARLOGDATAFMT_12BITPICIO;
		break;
	 case 11:
		if (format == SAMPLEFORMAT_VOID || format == SAMPLEFORMAT_UINT)
			guess = PIXARLOGDATAFMT_11BITLOG;
		break;
	 case 8:
		if (format == SAMPLEFORMAT_VOID || format == SAMPLEFORMAT_UINT)
			guess = PIXARLOGDATAFMT_8BIT;
		break;
	}

	return guess;
}

static int
PixarLogSetupDecode(TIFF* tif)
{
	TIFFDirectory *td = &tif->tif_dir;
	PixarLogState* sp = DecoderState(tif);
	static const char module[] = "PixarLogSetupDecode";

	assert(sp != NULL);

	/* Make sure no byte swapping happens on the data
	 * after decompression. */
	tif->tif_postdecode = _TIFFNoPostDecode;

	/* for some reason, we can't do this in TIFFInitPixarLog */

	sp->stride = (td->td_planarconfig == PLANARCONFIG_CONTIG ?
	    td->td_samplesperpixel : 1);
	sp->tbuf = (uint16 *) _TIFFmalloc(sp->stride *
		td->td_imagewidth * td->td_rowsperstrip * sizeof(uint16));
	if (sp->user_datafmt == PIXARLOGDATAFMT_UNKNOWN)
		sp->user_datafmt = PixarLogGuessDataFmt(td);
	if (sp->user_datafmt == PIXARLOGDATAFMT_UNKNOWN) {
		TIFFError(module,
			"PixarLog compression can't handle bits depth/data format combination (depth: %d)",
			td->td_bitspersample);
		return (0);
	}

	if (inflateInit(&sp->stream) != Z_OK) {
		TIFFError(module, "%s: %s", tif->tif_name, sp->stream.msg);
		return (0);
	} else {
		sp->state |= PLSTATE_INIT;
		return (1);
	}
}

/*
 * Setup state for decoding a strip.
 */
static int
PixarLogPreDecode(TIFF* tif, tsample_t s)
{
	TIFFDirectory *td = &tif->tif_dir;
	PixarLogState* sp = DecoderState(tif);

	(void) s;
	assert(sp != NULL);
	sp->stream.next_in = tif->tif_rawdata;
	sp->stream.avail_in = tif->tif_rawcc;
	return (inflateReset(&sp->stream) == Z_OK);
}

static int
PixarLogDecode(TIFF* tif, tidata_t op, tsize_t occ, tsample_t s)
{
	TIFFDirectory *td = &tif->tif_dir;
	PixarLogState* sp = DecoderState(tif);
	static const char module[] = "PixarLogDecode";
	int i, nsamples, llen;
	uint16 *up;

	switch (sp->user_datafmt) {
	case PIXARLOGDATAFMT_FLOAT:
		nsamples = occ / sizeof(float);	/* XXX float == 32 bits */
		break;
	case PIXARLOGDATAFMT_16BIT:
	case PIXARLOGDATAFMT_12BITPICIO:
	case PIXARLOGDATAFMT_11BITLOG:
		nsamples = occ / sizeof(uint16); /* XXX uint16 == 16 bits */
		break;
	case PIXARLOGDATAFMT_8BIT:
	case PIXARLOGDATAFMT_8BITABGR:
		nsamples = occ;
		break;
	default:
		TIFFError(tif->tif_name,
			"%d bit input not supported in PixarLog",
			td->td_bitspersample);
		return 0;
	}

	llen = sp->stride * td->td_imagewidth;

	(void) s;
	assert(sp != NULL);
	sp->stream.next_out = (unsigned char *) sp->tbuf;
	sp->stream.avail_out = nsamples * sizeof(uint16);
	do {
		int state = inflate(&sp->stream, Z_PARTIAL_FLUSH);
		if (state == Z_STREAM_END) {
			break;			/* XXX */
		}
		if (state == Z_DATA_ERROR) {
			TIFFError(module,
			    "%s: Decoding error at scanline %d, %s",
			    tif->tif_name, tif->tif_row, sp->stream.msg);
			if (inflateSync(&sp->stream) != Z_OK)
				return (0);
			continue;
		}
		if (state != Z_OK) {
			TIFFError(module, "%s: zlib error: %s",
			    tif->tif_name, sp->stream.msg);
			return (0);
		}
	} while (sp->stream.avail_out > 0);

	/* hopefully, we got all the bytes we needed */
	if (sp->stream.avail_out != 0) {
		TIFFError(module,
		    "%s: Not enough data at scanline %d (short %d bytes)",
		    tif->tif_name, tif->tif_row, sp->stream.avail_out);
		return (0);
	}

	up = sp->tbuf;
	/* Swap bytes in the data if from a different endian machine. */
	if (tif->tif_flags & TIFF_SWAB)
		TIFFSwabArrayOfShort(up, nsamples);

	for (i = 0; i < nsamples; i += llen, up += llen) {
		switch (sp->user_datafmt)  {
		case PIXARLOGDATAFMT_FLOAT:
			horizontalAccumulateF(up, llen, sp->stride,
					(float *)op, sp->ToLinearF);
			op += llen * sizeof(float);
			break;
		case PIXARLOGDATAFMT_16BIT:
			horizontalAccumulate16(up, llen, sp->stride,
					(uint16 *)op, sp->ToLinear16);
			op += llen * sizeof(uint16);
			break;
		case PIXARLOGDATAFMT_12BITPICIO:
			horizontalAccumulate12(up, llen, sp->stride,
					(int16 *)op, sp->ToLinearF);
			op += llen * sizeof(int16);
			break;
		case PIXARLOGDATAFMT_11BITLOG:
			horizontalAccumulate11(up, llen, sp->stride,
					(uint16 *)op);
			op += llen * sizeof(uint16);
			break;
		case PIXARLOGDATAFMT_8BIT:
			horizontalAccumulate8(up, llen, sp->stride,
					(unsigned char *)op, sp->ToLinear8);
			op += llen * sizeof(unsigned char);
			break;
		case PIXARLOGDATAFMT_8BITABGR:
			horizontalAccumulate8abgr(up, llen, sp->stride,
					(unsigned char *)op, sp->ToLinear8);
			op += llen * sizeof(unsigned char);
			break;
		default:
			TIFFError(tif->tif_name,
				  "PixarLogDecode: unsupported bits/sample: %d",
				  td->td_bitspersample);
			return (0);
		}
	}

	return (1);
}

static int
PixarLogSetupEncode(TIFF* tif)
{
	TIFFDirectory *td = &tif->tif_dir;
	PixarLogState* sp = EncoderState(tif);
	static const char module[] = "PixarLogSetupEncode";

	assert(sp != NULL);

	/* for some reason, we can't do this in TIFFInitPixarLog */

	sp->stride = (td->td_planarconfig == PLANARCONFIG_CONTIG ?
	    td->td_samplesperpixel : 1);
	sp->tbuf = (uint16 *) _TIFFmalloc(sp->stride *
		td->td_imagewidth * td->td_rowsperstrip * sizeof(uint16));
	if (sp->user_datafmt == PIXARLOGDATAFMT_UNKNOWN)
		sp->user_datafmt = PixarLogGuessDataFmt(td);
	if (sp->user_datafmt == PIXARLOGDATAFMT_UNKNOWN) {
		TIFFError(module, "PixarLog compression can't handle %d bit linear encodings", td->td_bitspersample);
		return (0);
	}

	if (deflateInit(&sp->stream, sp->quality) != Z_OK) {
		TIFFError(module, "%s: %s", tif->tif_name, sp->stream.msg);
		return (0);
	} else {
		sp->state |= PLSTATE_INIT;
		return (1);
	}
}

/*
 * Reset encoding state at the start of a strip.
 */
static int
PixarLogPreEncode(TIFF* tif, tsample_t s)
{
	TIFFDirectory *td = &tif->tif_dir;
	PixarLogState *sp = EncoderState(tif);

	(void) s;
	assert(sp != NULL);
	sp->stream.next_out = tif->tif_rawdata;
	sp->stream.avail_out = tif->tif_rawdatasize;
	return (deflateReset(&sp->stream) == Z_OK);
}

static void
horizontalDifferenceF(float *ip, int n, int stride, uint16 *wp, uint16 *FromLT2)
{

    register int  r1, g1, b1, a1, r2, g2, b2, a2, mask;
    register float  fltsize = Fltsize;

#define  CLAMP(v) ( (v<(float)0.)   ? 0				\
		  : (v<(float)2.)   ? FromLT2[(int)(v*fltsize)]	\
		  : (v>(float)24.2) ? 2047			\
		  : LogK1*log(v*LogK2) + 0.5 )

    mask = CODE_MASK;
    if (n >= stride) {
	if (stride == 3) {
	    r2 = wp[0] = CLAMP(ip[0]);  g2 = wp[1] = CLAMP(ip[1]);
	    b2 = wp[2] = CLAMP(ip[2]);
	    n -= 3;
	    while (n > 0) {
		n -= 3;
		wp += 3;
		ip += 3;
		r1 = CLAMP(ip[0]); wp[0] = (r1-r2) & mask; r2 = r1;
		g1 = CLAMP(ip[1]); wp[1] = (g1-g2) & mask; g2 = g1;
		b1 = CLAMP(ip[2]); wp[2] = (b1-b2) & mask; b2 = b1;
	    }
	} else if (stride == 4) {
	    r2 = wp[0] = CLAMP(ip[0]);  g2 = wp[1] = CLAMP(ip[1]);
	    b2 = wp[2] = CLAMP(ip[2]);  a2 = wp[3] = CLAMP(ip[3]);
	    n -= 4;
	    while (n > 0) {
		n -= 4;
		wp += 4;
		ip += 4;
		r1 = CLAMP(ip[0]); wp[0] = (r1-r2) & mask; r2 = r1;
		g1 = CLAMP(ip[1]); wp[1] = (g1-g2) & mask; g2 = g1;
		b1 = CLAMP(ip[2]); wp[2] = (b1-b2) & mask; b2 = b1;
		a1 = CLAMP(ip[3]); wp[3] = (a1-a2) & mask; a2 = a1;
	    }
	} else {
	    ip += n - 1;	/* point to last one */
	    wp += n - 1;	/* point to last one */
	    n -= stride;
	    while (n > 0) {
		REPEAT(stride, wp[0] = CLAMP(ip[0]);
				wp[stride] -= wp[0];
				wp[stride] &= mask;
				wp--; ip--)
		n -= stride;
	    }
	    REPEAT(stride, wp[0] = CLAMP(ip[0]); wp--; ip--)
	}
    }
}

static void
horizontalDifference16(unsigned short *ip, int n, int stride,
	unsigned short *wp, uint16 *From14)
{
    register int  r1, g1, b1, a1, r2, g2, b2, a2, mask;

/* assumption is unsigned pixel values */
#undef   CLAMP
#define  CLAMP(v) From14[(v) >> 2]

    mask = CODE_MASK;
    if (n >= stride) {
	if (stride == 3) {
	    r2 = wp[0] = CLAMP(ip[0]);  g2 = wp[1] = CLAMP(ip[1]);
	    b2 = wp[2] = CLAMP(ip[2]);
	    n -= 3;
	    while (n > 0) {
		n -= 3;
		wp += 3;
		ip += 3;
		r1 = CLAMP(ip[0]); wp[0] = (r1-r2) & mask; r2 = r1;
		g1 = CLAMP(ip[1]); wp[1] = (g1-g2) & mask; g2 = g1;
		b1 = CLAMP(ip[2]); wp[2] = (b1-b2) & mask; b2 = b1;
	    }
	} else if (stride == 4) {
	    r2 = wp[0] = CLAMP(ip[0]);  g2 = wp[1] = CLAMP(ip[1]);
	    b2 = wp[2] = CLAMP(ip[2]);  a2 = wp[3] = CLAMP(ip[3]);
	    n -= 4;
	    while (n > 0) {
		n -= 4;
		wp += 4;
		ip += 4;
		r1 = CLAMP(ip[0]); wp[0] = (r1-r2) & mask; r2 = r1;
		g1 = CLAMP(ip[1]); wp[1] = (g1-g2) & mask; g2 = g1;
		b1 = CLAMP(ip[2]); wp[2] = (b1-b2) & mask; b2 = b1;
		a1 = CLAMP(ip[3]); wp[3] = (a1-a2) & mask; a2 = a1;
	    }
	} else {
	    ip += n - 1;	/* point to last one */
	    wp += n - 1;	/* point to last one */
	    n -= stride;
	    while (n > 0) {
		REPEAT(stride, wp[0] = CLAMP(ip[0]);
				wp[stride] -= wp[0];
				wp[stride] &= mask;
				wp--; ip--)
		n -= stride;
	    }
	    REPEAT(stride, wp[0] = CLAMP(ip[0]); wp--; ip--)
	}
    }
}


static void
horizontalDifference8(unsigned char *ip, int n, int stride,
	unsigned short *wp, uint16 *From8)
{
    register int  r1, g1, b1, a1, r2, g2, b2, a2, mask;

#undef	 CLAMP
#define  CLAMP(v) (From8[(v)])

    mask = CODE_MASK;
    if (n >= stride) {
	if (stride == 3) {
	    r2 = wp[0] = CLAMP(ip[0]);  g2 = wp[1] = CLAMP(ip[1]);
	    b2 = wp[2] = CLAMP(ip[2]);
	    n -= 3;
	    while (n > 0) {
		n -= 3;
		r1 = CLAMP(ip[3]); wp[3] = (r1-r2) & mask; r2 = r1;
		g1 = CLAMP(ip[4]); wp[4] = (g1-g2) & mask; g2 = g1;
		b1 = CLAMP(ip[5]); wp[5] = (b1-b2) & mask; b2 = b1;
		wp += 3;
		ip += 3;
	    }
	} else if (stride == 4) {
	    r2 = wp[0] = CLAMP(ip[0]);  g2 = wp[1] = CLAMP(ip[1]);
	    b2 = wp[2] = CLAMP(ip[2]);  a2 = wp[3] = CLAMP(ip[3]);
	    n -= 4;
	    while (n > 0) {
		n -= 4;
		r1 = CLAMP(ip[4]); wp[4] = (r1-r2) & mask; r2 = r1;
		g1 = CLAMP(ip[5]); wp[5] = (g1-g2) & mask; g2 = g1;
		b1 = CLAMP(ip[6]); wp[6] = (b1-b2) & mask; b2 = b1;
		a1 = CLAMP(ip[7]); wp[7] = (a1-a2) & mask; a2 = a1;
		wp += 4;
		ip += 4;
	    }
	} else {
	    wp += n + stride - 1;	/* point to last one */
	    ip += n + stride - 1;	/* point to last one */
	    n -= stride;
	    while (n > 0) {
		REPEAT(stride, wp[0] = CLAMP(ip[0]);
				wp[stride] -= wp[0];
				wp[stride] &= mask;
				wp--; ip--)
		n -= stride;
	    }
	    REPEAT(stride, wp[0] = CLAMP(ip[0]); wp--; ip--)
	}
    }
}

/*
 * Encode a chunk of pixels.
 */
static int
PixarLogEncode(TIFF* tif, tidata_t bp, tsize_t cc, tsample_t s)
{
	TIFFDirectory *td = &tif->tif_dir;
	PixarLogState *sp = EncoderState(tif);
	static const char module[] = "PixarLogEncode";
	int 	i, n, llen;
	unsigned short * up;

	(void) s;

	switch (sp->user_datafmt) {
	case PIXARLOGDATAFMT_FLOAT:
		n = cc / sizeof(float);		/* XXX float == 32 bits */
		break;
	case PIXARLOGDATAFMT_16BIT:
	case PIXARLOGDATAFMT_12BITPICIO:
	case PIXARLOGDATAFMT_11BITLOG:
		n = cc / sizeof(uint16);	/* XXX uint16 == 16 bits */
		break;
	case PIXARLOGDATAFMT_8BIT:
	case PIXARLOGDATAFMT_8BITABGR:
		n = cc;
		break;
	default:
		TIFFError(tif->tif_name,
			"%d bit input not supported in PixarLog",
			td->td_bitspersample);
		return 0;
	}

	llen = sp->stride * td->td_imagewidth;

	for (i = 0, up = sp->tbuf; i < n; i += llen, up += llen) {
		switch (sp->user_datafmt)  {
		case PIXARLOGDATAFMT_FLOAT:
			horizontalDifferenceF((float *)bp, llen,
				sp->stride, up, sp->FromLT2);
			bp += llen * sizeof(float);
			break;
		case PIXARLOGDATAFMT_16BIT:
			horizontalDifference16((uint16 *)bp, llen,
				sp->stride, up, sp->From14);
			bp += llen * sizeof(uint16);
			break;
		case PIXARLOGDATAFMT_8BIT:
			horizontalDifference8((unsigned char *)bp, llen,
				sp->stride, up, sp->From8);
			bp += llen * sizeof(unsigned char);
			break;
		default:
			TIFFError(tif->tif_name,
				"%d bit input not supported in PixarLog",
				td->td_bitspersample);
			return 0;
		}
	}

	sp->stream.next_in = (unsigned char *) sp->tbuf;
	sp->stream.avail_in = n * sizeof(uint16);

	do {
		if (deflate(&sp->stream, Z_NO_FLUSH) != Z_OK) {
			TIFFError(module, "%s: Encoder error: %s",
			    tif->tif_name, sp->stream.msg);
			return (0);
		}
		if (sp->stream.avail_out == 0) {
			tif->tif_rawcc = tif->tif_rawdatasize;
			TIFFFlushData1(tif);
			sp->stream.next_out = tif->tif_rawdata;
			sp->stream.avail_out = tif->tif_rawdatasize;
		}
	} while (sp->stream.avail_in > 0);
	return (1);
}

/*
 * Finish off an encoded strip by flushing the last
 * string and tacking on an End Of Information code.
 */

static int
PixarLogPostEncode(TIFF* tif)
{
	PixarLogState *sp = EncoderState(tif);
	static const char module[] = "PixarLogPostEncode";
	int state;

	sp->stream.avail_in = 0;

	do {
		state = deflate(&sp->stream, Z_FINISH);
		switch (state) {
		case Z_STREAM_END:
		case Z_OK:
		    if (sp->stream.avail_out != tif->tif_rawdatasize) {
			    tif->tif_rawcc =
				tif->tif_rawdatasize - sp->stream.avail_out;
			    TIFFFlushData1(tif);
			    sp->stream.next_out = tif->tif_rawdata;
			    sp->stream.avail_out = tif->tif_rawdatasize;
		    }
		    break;
		default:
		    TIFFError(module, "%s: zlib error: %s",
			tif->tif_name, sp->stream.msg);
		    return (0);
		}
	} while (state != Z_STREAM_END);
	return (1);
}

static void
PixarLogClose(TIFF* tif)
{
	TIFFDirectory *td = &tif->tif_dir;

	/* In a really sneaky maneuver, on close, we covertly modify both
	 * bitspersample and sampleformat in the directory to indicate
	 * 8-bit linear.  This way, the decode "just works" even for
	 * readers that don't know about PixarLog, or how to set
	 * the PIXARLOGDATFMT pseudo-tag.
	 */
	td->td_bitspersample = 8;
	td->td_sampleformat = SAMPLEFORMAT_UINT;
}

static void
PixarLogCleanup(TIFF* tif)
{
	PixarLogState* sp = (PixarLogState*) tif->tif_data;

	if (sp) {
		if (sp->FromLT2) _TIFFfree(sp->FromLT2);
		if (sp->From14) _TIFFfree(sp->From14);
		if (sp->From8) _TIFFfree(sp->From8);
		if (sp->ToLinearF) _TIFFfree(sp->ToLinearF);
		if (sp->ToLinear16) _TIFFfree(sp->ToLinear16);
		if (sp->ToLinear8) _TIFFfree(sp->ToLinear8);
		if (sp->state&PLSTATE_INIT) {
			if (tif->tif_mode == O_RDONLY)
				inflateEnd(&sp->stream);
			else
				deflateEnd(&sp->stream);
		}
		if (sp->tbuf)
			_TIFFfree(sp->tbuf);
		_TIFFfree(sp);
		tif->tif_data = NULL;
	}
}

static int
PixarLogVSetField(TIFF* tif, ttag_t tag, va_list ap)
{
    PixarLogState *sp = (PixarLogState *)tif->tif_data;
    int result;
    static const char module[] = "PixarLogVSetField";

    switch (tag) {
     case TIFFTAG_PIXARLOGQUALITY:
		sp->quality = va_arg(ap, int);
		if (tif->tif_mode != O_RDONLY && (sp->state&PLSTATE_INIT)) {
			if (deflateParams(&sp->stream,
			    sp->quality, Z_DEFAULT_STRATEGY) != Z_OK) {
				TIFFError(module, "%s: zlib error: %s",
					tif->tif_name, sp->stream.msg);
				return (0);
			}
		}
		return (1);
     case TIFFTAG_PIXARLOGDATAFMT:
	sp->user_datafmt = va_arg(ap, int);
	/* Tweak the TIFF header so that the rest of libtiff knows what
	 * size of data will be passed between app and library, and
	 * assume that the app knows what it is doing and is not
	 * confused by these header manipulations...
	 */
	switch (sp->user_datafmt) {
	 case PIXARLOGDATAFMT_8BIT:
	 case PIXARLOGDATAFMT_8BITABGR:
	    TIFFSetField(tif, TIFFTAG_BITSPERSAMPLE, 8);
	    TIFFSetField(tif, TIFFTAG_SAMPLEFORMAT, SAMPLEFORMAT_UINT);
	    break;
	 case PIXARLOGDATAFMT_11BITLOG:
	    TIFFSetField(tif, TIFFTAG_BITSPERSAMPLE, 16);
	    TIFFSetField(tif, TIFFTAG_SAMPLEFORMAT, SAMPLEFORMAT_UINT);
	    break;
	 case PIXARLOGDATAFMT_12BITPICIO:
	    TIFFSetField(tif, TIFFTAG_BITSPERSAMPLE, 16);
	    TIFFSetField(tif, TIFFTAG_SAMPLEFORMAT, SAMPLEFORMAT_INT);
	    break;
	 case PIXARLOGDATAFMT_16BIT:
	    TIFFSetField(tif, TIFFTAG_BITSPERSAMPLE, 16);
	    TIFFSetField(tif, TIFFTAG_SAMPLEFORMAT, SAMPLEFORMAT_UINT);
	    break;
	 case PIXARLOGDATAFMT_FLOAT:
	    TIFFSetField(tif, TIFFTAG_BITSPERSAMPLE, 32);
	    TIFFSetField(tif, TIFFTAG_SAMPLEFORMAT, SAMPLEFORMAT_IEEEFP);
	    break;
	}
	/*
	 * Must recalculate sizes should bits/sample change.
	 */
	tif->tif_tilesize = TIFFTileSize(tif);
	tif->tif_scanlinesize = TIFFScanlineSize(tif);
	result = 1;		/* NB: pseudo tag */
	break;
     default:
	result = (*sp->vsetparent)(tif, tag, ap);
    }
    return (result);
}

static int
PixarLogVGetField(TIFF* tif, ttag_t tag, va_list ap)
{
    PixarLogState *sp = (PixarLogState *)tif->tif_data;

    switch (tag) {
     case TIFFTAG_PIXARLOGQUALITY:
	*va_arg(ap, int*) = sp->quality;
	break;
     case TIFFTAG_PIXARLOGDATAFMT:
	*va_arg(ap, int*) = sp->user_datafmt;
	break;
     default:
	return (*sp->vgetparent)(tif, tag, ap);
    }
    return (1);
}

static const TIFFFieldInfo pixarlogFieldInfo[] = {
    {TIFFTAG_PIXARLOGDATAFMT,0,0,TIFF_ANY,  FIELD_PSEUDO,FALSE,FALSE,""},
    {TIFFTAG_PIXARLOGQUALITY,0,0,TIFF_ANY,  FIELD_PSEUDO,FALSE,FALSE,""}
};

int
TIFFInitPixarLog(TIFF* tif, int scheme)
{
	PixarLogState* sp;

	assert(scheme == COMPRESSION_PIXARLOG);

	/*
	 * Allocate state block so tag methods have storage to record values.
	 */
	tif->tif_data = (tidata_t) _TIFFmalloc(sizeof (PixarLogState));
	if (tif->tif_data == NULL)
		goto bad;
	sp = (PixarLogState*) tif->tif_data;
	memset(sp, 0, sizeof (*sp));
	sp->stream.data_type = Z_BINARY;
	sp->user_datafmt = PIXARLOGDATAFMT_UNKNOWN;

	/*
	 * Install codec methods.
	 */
	tif->tif_setupdecode = PixarLogSetupDecode;
	tif->tif_predecode = PixarLogPreDecode;
	tif->tif_decoderow = PixarLogDecode;
	tif->tif_decodestrip = PixarLogDecode;
	tif->tif_decodetile = PixarLogDecode;
	tif->tif_setupencode = PixarLogSetupEncode;
	tif->tif_preencode = PixarLogPreEncode;
	tif->tif_postencode = PixarLogPostEncode;
	tif->tif_encoderow = PixarLogEncode;
	tif->tif_encodestrip = PixarLogEncode;
	tif->tif_encodetile = PixarLogEncode;
	tif->tif_close = PixarLogClose;
	tif->tif_cleanup = PixarLogCleanup;

	/* Override SetField so we can handle our private pseudo-tag */
	_TIFFMergeFieldInfo(tif, pixarlogFieldInfo, N(pixarlogFieldInfo));
	sp->vgetparent = tif->tif_vgetfield;
	tif->tif_vgetfield = PixarLogVGetField;   /* hook for codec tags */
	sp->vsetparent = tif->tif_vsetfield;
	tif->tif_vsetfield = PixarLogVSetField;   /* hook for codec tags */

	/* Default values for codec-specific fields */
	sp->quality = Z_DEFAULT_COMPRESSION; /* default comp. level */
	sp->state = 0;

	/* we don't wish to use the predictor,
	 * the default is none, which predictor value 1
	 */
	(void) TIFFPredictorInit(tif);

	/*
	 * build the companding tables
	 */
	PixarLogMakeTables(sp);

	return (1);
bad:
	TIFFError("TIFFInitPixarLog", "No space for PixarLog state block");
	return (0);
}
#endif /* PIXARLOG_SUPPORT */