Home | History | Annotate | Download | only in libtiff
      1 /* $Id: tif_pixarlog.c,v 1.39 2012-12-10 17:27:13 tgl Exp $ */
      2 
      3 /*
      4  * Copyright (c) 1996-1997 Sam Leffler
      5  * Copyright (c) 1996 Pixar
      6  *
      7  * Permission to use, copy, modify, distribute, and sell this software and
      8  * its documentation for any purpose is hereby granted without fee, provided
      9  * that (i) the above copyright notices and this permission notice appear in
     10  * all copies of the software and related documentation, and (ii) the names of
     11  * Pixar, Sam Leffler and Silicon Graphics may not be used in any advertising or
     12  * publicity relating to the software without the specific, prior written
     13  * permission of Pixar, Sam Leffler and Silicon Graphics.
     14  *
     15  * THE SOFTWARE IS PROVIDED "AS-IS" AND WITHOUT WARRANTY OF ANY KIND,
     16  * EXPRESS, IMPLIED OR OTHERWISE, INCLUDING WITHOUT LIMITATION, ANY
     17  * WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
     18  *
     19  * IN NO EVENT SHALL PIXAR, SAM LEFFLER OR SILICON GRAPHICS BE LIABLE FOR
     20  * ANY SPECIAL, INCIDENTAL, INDIRECT OR CONSEQUENTIAL DAMAGES OF ANY KIND,
     21  * OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS,
     22  * WHETHER OR NOT ADVISED OF THE POSSIBILITY OF DAMAGE, AND ON ANY THEORY OF
     23  * LIABILITY, ARISING OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE
     24  * OF THIS SOFTWARE.
     25  */
     26 
     27 #include "tiffiop.h"
     28 #ifdef PIXARLOG_SUPPORT
     29 
     30 /*
     31  * TIFF Library.
     32  * PixarLog Compression Support
     33  *
     34  * Contributed by Dan McCoy.
     35  *
     36  * PixarLog film support uses the TIFF library to store companded
     37  * 11 bit values into a tiff file, which are compressed using the
     38  * zip compressor.
     39  *
     40  * The codec can take as input and produce as output 32-bit IEEE float values
     41  * as well as 16-bit or 8-bit unsigned integer values.
     42  *
     43  * On writing any of the above are converted into the internal
     44  * 11-bit log format.   In the case of  8 and 16 bit values, the
     45  * input is assumed to be unsigned linear color values that represent
     46  * the range 0-1.  In the case of IEEE values, the 0-1 range is assumed to
     47  * be the normal linear color range, in addition over 1 values are
     48  * accepted up to a value of about 25.0 to encode "hot" hightlights and such.
     49  * The encoding is lossless for 8-bit values, slightly lossy for the
     50  * other bit depths.  The actual color precision should be better
     51  * than the human eye can perceive with extra room to allow for
     52  * error introduced by further image computation.  As with any quantized
     53  * color format, it is possible to perform image calculations which
     54  * expose the quantization error. This format should certainly be less
     55  * susceptable to such errors than standard 8-bit encodings, but more
     56  * susceptable than straight 16-bit or 32-bit encodings.
     57  *
     58  * On reading the internal format is converted to the desired output format.
     59  * The program can request which format it desires by setting the internal
     60  * pseudo tag TIFFTAG_PIXARLOGDATAFMT to one of these possible values:
     61  *  PIXARLOGDATAFMT_FLOAT     = provide IEEE float values.
     62  *  PIXARLOGDATAFMT_16BIT     = provide unsigned 16-bit integer values
     63  *  PIXARLOGDATAFMT_8BIT      = provide unsigned 8-bit integer values
     64  *
     65  * alternately PIXARLOGDATAFMT_8BITABGR provides unsigned 8-bit integer
     66  * values with the difference that if there are exactly three or four channels
     67  * (rgb or rgba) it swaps the channel order (bgr or abgr).
     68  *
     69  * PIXARLOGDATAFMT_11BITLOG provides the internal encoding directly
     70  * packed in 16-bit values.   However no tools are supplied for interpreting
     71  * these values.
     72  *
     73  * "hot" (over 1.0) areas written in floating point get clamped to
     74  * 1.0 in the integer data types.
     75  *
     76  * When the file is closed after writing, the bit depth and sample format
     77  * are set always to appear as if 8-bit data has been written into it.
     78  * That way a naive program unaware of the particulars of the encoding
     79  * gets the format it is most likely able to handle.
     80  *
     81  * The codec does it's own horizontal differencing step on the coded
     82  * values so the libraries predictor stuff should be turned off.
     83  * The codec also handle byte swapping the encoded values as necessary
     84  * since the library does not have the information necessary
     85  * to know the bit depth of the raw unencoded buffer.
     86  *
     87  * NOTE: This decoder does not appear to update tif_rawcp, and tif_rawcc.
     88  * This can cause problems with the implementation of CHUNKY_STRIP_READ_SUPPORT
     89  * as noted in http://trac.osgeo.org/gdal/ticket/3894.   FrankW - Jan'11
     90  */
     91 
     92 #include "tif_predict.h"
     93 #include "../zlib_v128/zlib.h"
     94 
     95 #include <stdio.h>
     96 #include <stdlib.h>
     97 #include <math.h>
     98 
     99 /* Tables for converting to/from 11 bit coded values */
    100 
    101 #define  TSIZE	 2048		/* decode table size (11-bit tokens) */
    102 #define  TSIZEP1 2049		/* Plus one for slop */
    103 #define  ONE	 1250		/* token value of 1.0 exactly */
    104 #define  RATIO	 1.004		/* nominal ratio for log part */
    105 
    106 #define CODE_MASK 0x7ff         /* 11 bits. */
    107 
    108 static float  Fltsize;
    109 static float  LogK1, LogK2;
    110 
    111 #define REPEAT(n, op)   { int i; i=n; do { i--; op; } while (i>0); }
    112 
    113 static void
    114 horizontalAccumulateF(uint16 *wp, int n, int stride, float *op,
    115 	float *ToLinearF)
    116 {
    117     register unsigned int  cr, cg, cb, ca, mask;
    118     register float  t0, t1, t2, t3;
    119 
    120     if (n >= stride) {
    121 	mask = CODE_MASK;
    122 	if (stride == 3) {
    123 	    t0 = ToLinearF[cr = (wp[0] & mask)];
    124 	    t1 = ToLinearF[cg = (wp[1] & mask)];
    125 	    t2 = ToLinearF[cb = (wp[2] & mask)];
    126 	    op[0] = t0;
    127 	    op[1] = t1;
    128 	    op[2] = t2;
    129 	    n -= 3;
    130 	    while (n > 0) {
    131 		wp += 3;
    132 		op += 3;
    133 		n -= 3;
    134 		t0 = ToLinearF[(cr += wp[0]) & mask];
    135 		t1 = ToLinearF[(cg += wp[1]) & mask];
    136 		t2 = ToLinearF[(cb += wp[2]) & mask];
    137 		op[0] = t0;
    138 		op[1] = t1;
    139 		op[2] = t2;
    140 	    }
    141 	} else if (stride == 4) {
    142 	    t0 = ToLinearF[cr = (wp[0] & mask)];
    143 	    t1 = ToLinearF[cg = (wp[1] & mask)];
    144 	    t2 = ToLinearF[cb = (wp[2] & mask)];
    145 	    t3 = ToLinearF[ca = (wp[3] & mask)];
    146 	    op[0] = t0;
    147 	    op[1] = t1;
    148 	    op[2] = t2;
    149 	    op[3] = t3;
    150 	    n -= 4;
    151 	    while (n > 0) {
    152 		wp += 4;
    153 		op += 4;
    154 		n -= 4;
    155 		t0 = ToLinearF[(cr += wp[0]) & mask];
    156 		t1 = ToLinearF[(cg += wp[1]) & mask];
    157 		t2 = ToLinearF[(cb += wp[2]) & mask];
    158 		t3 = ToLinearF[(ca += wp[3]) & mask];
    159 		op[0] = t0;
    160 		op[1] = t1;
    161 		op[2] = t2;
    162 		op[3] = t3;
    163 	    }
    164 	} else {
    165 	    REPEAT(stride, *op = ToLinearF[*wp&mask]; wp++; op++)
    166 	    n -= stride;
    167 	    while (n > 0) {
    168 		REPEAT(stride,
    169 		    wp[stride] += *wp; *op = ToLinearF[*wp&mask]; wp++; op++)
    170 		n -= stride;
    171 	    }
    172 	}
    173     }
    174 }
    175 
    176 static void
    177 horizontalAccumulate12(uint16 *wp, int n, int stride, int16 *op,
    178 	float *ToLinearF)
    179 {
    180     register unsigned int  cr, cg, cb, ca, mask;
    181     register float  t0, t1, t2, t3;
    182 
    183 #define SCALE12 2048.0F
    184 #define CLAMP12(t) (((t) < 3071) ? (uint16) (t) : 3071)
    185 
    186     if (n >= stride) {
    187 	mask = CODE_MASK;
    188 	if (stride == 3) {
    189 	    t0 = ToLinearF[cr = (wp[0] & mask)] * SCALE12;
    190 	    t1 = ToLinearF[cg = (wp[1] & mask)] * SCALE12;
    191 	    t2 = ToLinearF[cb = (wp[2] & mask)] * SCALE12;
    192 	    op[0] = CLAMP12(t0);
    193 	    op[1] = CLAMP12(t1);
    194 	    op[2] = CLAMP12(t2);
    195 	    n -= 3;
    196 	    while (n > 0) {
    197 		wp += 3;
    198 		op += 3;
    199 		n -= 3;
    200 		t0 = ToLinearF[(cr += wp[0]) & mask] * SCALE12;
    201 		t1 = ToLinearF[(cg += wp[1]) & mask] * SCALE12;
    202 		t2 = ToLinearF[(cb += wp[2]) & mask] * SCALE12;
    203 		op[0] = CLAMP12(t0);
    204 		op[1] = CLAMP12(t1);
    205 		op[2] = CLAMP12(t2);
    206 	    }
    207 	} else if (stride == 4) {
    208 	    t0 = ToLinearF[cr = (wp[0] & mask)] * SCALE12;
    209 	    t1 = ToLinearF[cg = (wp[1] & mask)] * SCALE12;
    210 	    t2 = ToLinearF[cb = (wp[2] & mask)] * SCALE12;
    211 	    t3 = ToLinearF[ca = (wp[3] & mask)] * SCALE12;
    212 	    op[0] = CLAMP12(t0);
    213 	    op[1] = CLAMP12(t1);
    214 	    op[2] = CLAMP12(t2);
    215 	    op[3] = CLAMP12(t3);
    216 	    n -= 4;
    217 	    while (n > 0) {
    218 		wp += 4;
    219 		op += 4;
    220 		n -= 4;
    221 		t0 = ToLinearF[(cr += wp[0]) & mask] * SCALE12;
    222 		t1 = ToLinearF[(cg += wp[1]) & mask] * SCALE12;
    223 		t2 = ToLinearF[(cb += wp[2]) & mask] * SCALE12;
    224 		t3 = ToLinearF[(ca += wp[3]) & mask] * SCALE12;
    225 		op[0] = CLAMP12(t0);
    226 		op[1] = CLAMP12(t1);
    227 		op[2] = CLAMP12(t2);
    228 		op[3] = CLAMP12(t3);
    229 	    }
    230 	} else {
    231 	    REPEAT(stride, t0 = ToLinearF[*wp&mask] * SCALE12;
    232                            *op = CLAMP12(t0); wp++; op++)
    233 	    n -= stride;
    234 	    while (n > 0) {
    235 		REPEAT(stride,
    236 		    wp[stride] += *wp; t0 = ToLinearF[wp[stride]&mask]*SCALE12;
    237 		    *op = CLAMP12(t0);  wp++; op++)
    238 		n -= stride;
    239 	    }
    240 	}
    241     }
    242 }
    243 
    244 static void
    245 horizontalAccumulate16(uint16 *wp, int n, int stride, uint16 *op,
    246 	uint16 *ToLinear16)
    247 {
    248     register unsigned int  cr, cg, cb, ca, mask;
    249 
    250     if (n >= stride) {
    251 	mask = CODE_MASK;
    252 	if (stride == 3) {
    253 	    op[0] = ToLinear16[cr = (wp[0] & mask)];
    254 	    op[1] = ToLinear16[cg = (wp[1] & mask)];
    255 	    op[2] = ToLinear16[cb = (wp[2] & mask)];
    256 	    n -= 3;
    257 	    while (n > 0) {
    258 		wp += 3;
    259 		op += 3;
    260 		n -= 3;
    261 		op[0] = ToLinear16[(cr += wp[0]) & mask];
    262 		op[1] = ToLinear16[(cg += wp[1]) & mask];
    263 		op[2] = ToLinear16[(cb += wp[2]) & mask];
    264 	    }
    265 	} else if (stride == 4) {
    266 	    op[0] = ToLinear16[cr = (wp[0] & mask)];
    267 	    op[1] = ToLinear16[cg = (wp[1] & mask)];
    268 	    op[2] = ToLinear16[cb = (wp[2] & mask)];
    269 	    op[3] = ToLinear16[ca = (wp[3] & mask)];
    270 	    n -= 4;
    271 	    while (n > 0) {
    272 		wp += 4;
    273 		op += 4;
    274 		n -= 4;
    275 		op[0] = ToLinear16[(cr += wp[0]) & mask];
    276 		op[1] = ToLinear16[(cg += wp[1]) & mask];
    277 		op[2] = ToLinear16[(cb += wp[2]) & mask];
    278 		op[3] = ToLinear16[(ca += wp[3]) & mask];
    279 	    }
    280 	} else {
    281 	    REPEAT(stride, *op = ToLinear16[*wp&mask]; wp++; op++)
    282 	    n -= stride;
    283 	    while (n > 0) {
    284 		REPEAT(stride,
    285 		    wp[stride] += *wp; *op = ToLinear16[*wp&mask]; wp++; op++)
    286 		n -= stride;
    287 	    }
    288 	}
    289     }
    290 }
    291 
    292 /*
    293  * Returns the log encoded 11-bit values with the horizontal
    294  * differencing undone.
    295  */
    296 static void
    297 horizontalAccumulate11(uint16 *wp, int n, int stride, uint16 *op)
    298 {
    299     register unsigned int  cr, cg, cb, ca, mask;
    300 
    301     if (n >= stride) {
    302 	mask = CODE_MASK;
    303 	if (stride == 3) {
    304 	    op[0] = cr = wp[0];  op[1] = cg = wp[1];  op[2] = cb = wp[2];
    305 	    n -= 3;
    306 	    while (n > 0) {
    307 		wp += 3;
    308 		op += 3;
    309 		n -= 3;
    310 		op[0] = (cr += wp[0]) & mask;
    311 		op[1] = (cg += wp[1]) & mask;
    312 		op[2] = (cb += wp[2]) & mask;
    313 	    }
    314 	} else if (stride == 4) {
    315 	    op[0] = cr = wp[0];  op[1] = cg = wp[1];
    316 	    op[2] = cb = wp[2];  op[3] = ca = wp[3];
    317 	    n -= 4;
    318 	    while (n > 0) {
    319 		wp += 4;
    320 		op += 4;
    321 		n -= 4;
    322 		op[0] = (cr += wp[0]) & mask;
    323 		op[1] = (cg += wp[1]) & mask;
    324 		op[2] = (cb += wp[2]) & mask;
    325 		op[3] = (ca += wp[3]) & mask;
    326 	    }
    327 	} else {
    328 	    REPEAT(stride, *op = *wp&mask; wp++; op++)
    329 	    n -= stride;
    330 	    while (n > 0) {
    331 		REPEAT(stride,
    332 		    wp[stride] += *wp; *op = *wp&mask; wp++; op++)
    333 		n -= stride;
    334 	    }
    335 	}
    336     }
    337 }
    338 
    339 static void
    340 horizontalAccumulate8(uint16 *wp, int n, int stride, unsigned char *op,
    341 	unsigned char *ToLinear8)
    342 {
    343     register unsigned int  cr, cg, cb, ca, mask;
    344 
    345     if (n >= stride) {
    346 	mask = CODE_MASK;
    347 	if (stride == 3) {
    348 	    op[0] = ToLinear8[cr = (wp[0] & mask)];
    349 	    op[1] = ToLinear8[cg = (wp[1] & mask)];
    350 	    op[2] = ToLinear8[cb = (wp[2] & mask)];
    351 	    n -= 3;
    352 	    while (n > 0) {
    353 		n -= 3;
    354 		wp += 3;
    355 		op += 3;
    356 		op[0] = ToLinear8[(cr += wp[0]) & mask];
    357 		op[1] = ToLinear8[(cg += wp[1]) & mask];
    358 		op[2] = ToLinear8[(cb += wp[2]) & mask];
    359 	    }
    360 	} else if (stride == 4) {
    361 	    op[0] = ToLinear8[cr = (wp[0] & mask)];
    362 	    op[1] = ToLinear8[cg = (wp[1] & mask)];
    363 	    op[2] = ToLinear8[cb = (wp[2] & mask)];
    364 	    op[3] = ToLinear8[ca = (wp[3] & mask)];
    365 	    n -= 4;
    366 	    while (n > 0) {
    367 		n -= 4;
    368 		wp += 4;
    369 		op += 4;
    370 		op[0] = ToLinear8[(cr += wp[0]) & mask];
    371 		op[1] = ToLinear8[(cg += wp[1]) & mask];
    372 		op[2] = ToLinear8[(cb += wp[2]) & mask];
    373 		op[3] = ToLinear8[(ca += wp[3]) & mask];
    374 	    }
    375 	} else {
    376 	    REPEAT(stride, *op = ToLinear8[*wp&mask]; wp++; op++)
    377 	    n -= stride;
    378 	    while (n > 0) {
    379 		REPEAT(stride,
    380 		    wp[stride] += *wp; *op = ToLinear8[*wp&mask]; wp++; op++)
    381 		n -= stride;
    382 	    }
    383 	}
    384     }
    385 }
    386 
    387 
    388 static void
    389 horizontalAccumulate8abgr(uint16 *wp, int n, int stride, unsigned char *op,
    390 	unsigned char *ToLinear8)
    391 {
    392     register unsigned int  cr, cg, cb, ca, mask;
    393     register unsigned char  t0, t1, t2, t3;
    394 
    395     if (n >= stride) {
    396 	mask = CODE_MASK;
    397 	if (stride == 3) {
    398 	    op[0] = 0;
    399 	    t1 = ToLinear8[cb = (wp[2] & mask)];
    400 	    t2 = ToLinear8[cg = (wp[1] & mask)];
    401 	    t3 = ToLinear8[cr = (wp[0] & mask)];
    402 	    op[1] = t1;
    403 	    op[2] = t2;
    404 	    op[3] = t3;
    405 	    n -= 3;
    406 	    while (n > 0) {
    407 		n -= 3;
    408 		wp += 3;
    409 		op += 4;
    410 		op[0] = 0;
    411 		t1 = ToLinear8[(cb += wp[2]) & mask];
    412 		t2 = ToLinear8[(cg += wp[1]) & mask];
    413 		t3 = ToLinear8[(cr += wp[0]) & mask];
    414 		op[1] = t1;
    415 		op[2] = t2;
    416 		op[3] = t3;
    417 	    }
    418 	} else if (stride == 4) {
    419 	    t0 = ToLinear8[ca = (wp[3] & mask)];
    420 	    t1 = ToLinear8[cb = (wp[2] & mask)];
    421 	    t2 = ToLinear8[cg = (wp[1] & mask)];
    422 	    t3 = ToLinear8[cr = (wp[0] & mask)];
    423 	    op[0] = t0;
    424 	    op[1] = t1;
    425 	    op[2] = t2;
    426 	    op[3] = t3;
    427 	    n -= 4;
    428 	    while (n > 0) {
    429 		n -= 4;
    430 		wp += 4;
    431 		op += 4;
    432 		t0 = ToLinear8[(ca += wp[3]) & mask];
    433 		t1 = ToLinear8[(cb += wp[2]) & mask];
    434 		t2 = ToLinear8[(cg += wp[1]) & mask];
    435 		t3 = ToLinear8[(cr += wp[0]) & mask];
    436 		op[0] = t0;
    437 		op[1] = t1;
    438 		op[2] = t2;
    439 		op[3] = t3;
    440 	    }
    441 	} else {
    442 	    REPEAT(stride, *op = ToLinear8[*wp&mask]; wp++; op++)
    443 	    n -= stride;
    444 	    while (n > 0) {
    445 		REPEAT(stride,
    446 		    wp[stride] += *wp; *op = ToLinear8[*wp&mask]; wp++; op++)
    447 		n -= stride;
    448 	    }
    449 	}
    450     }
    451 }
    452 
    453 /*
    454  * State block for each open TIFF
    455  * file using PixarLog compression/decompression.
    456  */
    457 typedef	struct {
    458 	TIFFPredictorState	predict;
    459 	z_stream		stream;
    460 	uint16			*tbuf;
    461 	uint16			stride;
    462 	int			state;
    463 	int			user_datafmt;
    464 	int			quality;
    465 #define PLSTATE_INIT 1
    466 
    467 	TIFFVSetMethod		vgetparent;	/* super-class method */
    468 	TIFFVSetMethod		vsetparent;	/* super-class method */
    469 
    470 	float *ToLinearF;
    471 	uint16 *ToLinear16;
    472 	unsigned char *ToLinear8;
    473 	uint16  *FromLT2;
    474 	uint16  *From14; /* Really for 16-bit data, but we shift down 2 */
    475 	uint16  *From8;
    476 
    477 } PixarLogState;
    478 
    479 static int
    480 PixarLogMakeTables(PixarLogState *sp)
    481 {
    482 
    483 /*
    484  *    We make several tables here to convert between various external
    485  *    representations (float, 16-bit, and 8-bit) and the internal
    486  *    11-bit companded representation.  The 11-bit representation has two
    487  *    distinct regions.  A linear bottom end up through .018316 in steps
    488  *    of about .000073, and a region of constant ratio up to about 25.
    489  *    These floating point numbers are stored in the main table ToLinearF.
    490  *    All other tables are derived from this one.  The tables (and the
    491  *    ratios) are continuous at the internal seam.
    492  */
    493 
    494     int  nlin, lt2size;
    495     int  i, j;
    496     double  b, c, linstep, v;
    497     float *ToLinearF;
    498     uint16 *ToLinear16;
    499     unsigned char *ToLinear8;
    500     uint16  *FromLT2;
    501     uint16  *From14; /* Really for 16-bit data, but we shift down 2 */
    502     uint16  *From8;
    503 
    504     c = log(RATIO);
    505     nlin = (int)(1./c);	/* nlin must be an integer */
    506     c = 1./nlin;
    507     b = exp(-c*ONE);	/* multiplicative scale factor [b*exp(c*ONE) = 1] */
    508     linstep = b*c*exp(1.);
    509 
    510     LogK1 = (float)(1./c);	/* if (v >= 2)  token = k1*log(v*k2) */
    511     LogK2 = (float)(1./b);
    512     lt2size = (int)(2./linstep) + 1;
    513     FromLT2 = (uint16 *)_TIFFmalloc(lt2size*sizeof(uint16));
    514     From14 = (uint16 *)_TIFFmalloc(16384*sizeof(uint16));
    515     From8 = (uint16 *)_TIFFmalloc(256*sizeof(uint16));
    516     ToLinearF = (float *)_TIFFmalloc(TSIZEP1 * sizeof(float));
    517     ToLinear16 = (uint16 *)_TIFFmalloc(TSIZEP1 * sizeof(uint16));
    518     ToLinear8 = (unsigned char *)_TIFFmalloc(TSIZEP1 * sizeof(unsigned char));
    519     if (FromLT2 == NULL || From14  == NULL || From8   == NULL ||
    520 	 ToLinearF == NULL || ToLinear16 == NULL || ToLinear8 == NULL) {
    521 	if (FromLT2) _TIFFfree(FromLT2);
    522 	if (From14) _TIFFfree(From14);
    523 	if (From8) _TIFFfree(From8);
    524 	if (ToLinearF) _TIFFfree(ToLinearF);
    525 	if (ToLinear16) _TIFFfree(ToLinear16);
    526 	if (ToLinear8) _TIFFfree(ToLinear8);
    527 	sp->FromLT2 = NULL;
    528 	sp->From14 = NULL;
    529 	sp->From8 = NULL;
    530 	sp->ToLinearF = NULL;
    531 	sp->ToLinear16 = NULL;
    532 	sp->ToLinear8 = NULL;
    533 	return 0;
    534     }
    535 
    536     j = 0;
    537 
    538     for (i = 0; i < nlin; i++)  {
    539 	v = i * linstep;
    540 	ToLinearF[j++] = (float)v;
    541     }
    542 
    543     for (i = nlin; i < TSIZE; i++)
    544 	ToLinearF[j++] = (float)(b*exp(c*i));
    545 
    546     ToLinearF[2048] = ToLinearF[2047];
    547 
    548     for (i = 0; i < TSIZEP1; i++)  {
    549 	v = ToLinearF[i]*65535.0 + 0.5;
    550 	ToLinear16[i] = (v > 65535.0) ? 65535 : (uint16)v;
    551 	v = ToLinearF[i]*255.0  + 0.5;
    552 	ToLinear8[i]  = (v > 255.0) ? 255 : (unsigned char)v;
    553     }
    554 
    555     j = 0;
    556     for (i = 0; i < lt2size; i++)  {
    557 	if ((i*linstep)*(i*linstep) > ToLinearF[j]*ToLinearF[j+1])
    558 	    j++;
    559 	FromLT2[i] = j;
    560     }
    561 
    562     /*
    563      * Since we lose info anyway on 16-bit data, we set up a 14-bit
    564      * table and shift 16-bit values down two bits on input.
    565      * saves a little table space.
    566      */
    567     j = 0;
    568     for (i = 0; i < 16384; i++)  {
    569 	while ((i/16383.)*(i/16383.) > ToLinearF[j]*ToLinearF[j+1])
    570 	    j++;
    571 	From14[i] = j;
    572     }
    573 
    574     j = 0;
    575     for (i = 0; i < 256; i++)  {
    576 	while ((i/255.)*(i/255.) > ToLinearF[j]*ToLinearF[j+1])
    577 	    j++;
    578 	From8[i] = j;
    579     }
    580 
    581     Fltsize = (float)(lt2size/2);
    582 
    583     sp->ToLinearF = ToLinearF;
    584     sp->ToLinear16 = ToLinear16;
    585     sp->ToLinear8 = ToLinear8;
    586     sp->FromLT2 = FromLT2;
    587     sp->From14 = From14;
    588     sp->From8 = From8;
    589 
    590     return 1;
    591 }
    592 
    593 #define DecoderState(tif)	((PixarLogState*) (tif)->tif_data)
    594 #define EncoderState(tif)	((PixarLogState*) (tif)->tif_data)
    595 
    596 static int PixarLogEncode(TIFF* tif, uint8* bp, tmsize_t cc, uint16 s);
    597 static int PixarLogDecode(TIFF* tif, uint8* op, tmsize_t occ, uint16 s);
    598 
    599 #define PIXARLOGDATAFMT_UNKNOWN	-1
    600 
    601 static int
    602 PixarLogGuessDataFmt(TIFFDirectory *td)
    603 {
    604 	int guess = PIXARLOGDATAFMT_UNKNOWN;
    605 	int format = td->td_sampleformat;
    606 
    607 	/* If the user didn't tell us his datafmt,
    608 	 * take our best guess from the bitspersample.
    609 	 */
    610 	switch (td->td_bitspersample) {
    611 	 case 32:
    612 		if (format == SAMPLEFORMAT_IEEEFP)
    613 			guess = PIXARLOGDATAFMT_FLOAT;
    614 		break;
    615 	 case 16:
    616 		if (format == SAMPLEFORMAT_VOID || format == SAMPLEFORMAT_UINT)
    617 			guess = PIXARLOGDATAFMT_16BIT;
    618 		break;
    619 	 case 12:
    620 		if (format == SAMPLEFORMAT_VOID || format == SAMPLEFORMAT_INT)
    621 			guess = PIXARLOGDATAFMT_12BITPICIO;
    622 		break;
    623 	 case 11:
    624 		if (format == SAMPLEFORMAT_VOID || format == SAMPLEFORMAT_UINT)
    625 			guess = PIXARLOGDATAFMT_11BITLOG;
    626 		break;
    627 	 case 8:
    628 		if (format == SAMPLEFORMAT_VOID || format == SAMPLEFORMAT_UINT)
    629 			guess = PIXARLOGDATAFMT_8BIT;
    630 		break;
    631 	}
    632 
    633 	return guess;
    634 }
    635 
    636 static tmsize_t
    637 multiply_ms(tmsize_t m1, tmsize_t m2)
    638 {
    639 	tmsize_t bytes = m1 * m2;
    640 
    641 	if (m1 && bytes / m1 != m2)
    642 		bytes = 0;
    643 
    644 	return bytes;
    645 }
    646 
    647 static tmsize_t
    648 add_ms(tmsize_t m1, tmsize_t m2)
    649 {
    650 	tmsize_t bytes = m1 + m2;
    651 
    652 	/* if either input is zero, assume overflow already occurred */
    653 	if (m1 == 0 || m2 == 0)
    654 		bytes = 0;
    655 	else if (bytes <= m1 || bytes <= m2)
    656 		bytes = 0;
    657 
    658 	return bytes;
    659 }
    660 
    661 static int
    662 PixarLogFixupTags(TIFF* tif)
    663 {
    664 	(void) tif;
    665 	return (1);
    666 }
    667 
    668 static int
    669 PixarLogSetupDecode(TIFF* tif)
    670 {
    671 	static const char module[] = "PixarLogSetupDecode";
    672 	TIFFDirectory *td = &tif->tif_dir;
    673 	PixarLogState* sp = DecoderState(tif);
    674 	tmsize_t tbuf_size;
    675 
    676 	assert(sp != NULL);
    677 
    678 	/* Make sure no byte swapping happens on the data
    679 	 * after decompression. */
    680 	tif->tif_postdecode = _TIFFNoPostDecode;
    681 
    682 	/* for some reason, we can't do this in TIFFInitPixarLog */
    683 
    684 	sp->stride = (td->td_planarconfig == PLANARCONFIG_CONTIG ?
    685 	    td->td_samplesperpixel : 1);
    686 	tbuf_size = multiply_ms(multiply_ms(multiply_ms(sp->stride, td->td_imagewidth),
    687 				      td->td_rowsperstrip), sizeof(uint16));
    688 	/* add one more stride in case input ends mid-stride */
    689 	tbuf_size = add_ms(tbuf_size, sizeof(uint16) * sp->stride);
    690 	if (tbuf_size == 0)
    691 		return (0);   /* TODO: this is an error return without error report through TIFFErrorExt */
    692 	sp->tbuf = (uint16 *) _TIFFmalloc(tbuf_size);
    693 	if (sp->tbuf == NULL)
    694 		return (0);
    695 	if (sp->user_datafmt == PIXARLOGDATAFMT_UNKNOWN)
    696 		sp->user_datafmt = PixarLogGuessDataFmt(td);
    697 	if (sp->user_datafmt == PIXARLOGDATAFMT_UNKNOWN) {
    698 		TIFFErrorExt(tif->tif_clientdata, module,
    699 			"PixarLog compression can't handle bits depth/data format combination (depth: %d)",
    700 			td->td_bitspersample);
    701 		return (0);
    702 	}
    703 
    704 	if (inflateInit(&sp->stream) != Z_OK) {
    705 		TIFFErrorExt(tif->tif_clientdata, module, "%s", sp->stream.msg);
    706 		return (0);
    707 	} else {
    708 		sp->state |= PLSTATE_INIT;
    709 		return (1);
    710 	}
    711 }
    712 
    713 /*
    714  * Setup state for decoding a strip.
    715  */
    716 static int
    717 PixarLogPreDecode(TIFF* tif, uint16 s)
    718 {
    719 	static const char module[] = "PixarLogPreDecode";
    720 	PixarLogState* sp = DecoderState(tif);
    721 
    722 	(void) s;
    723 	assert(sp != NULL);
    724 	sp->stream.next_in = tif->tif_rawdata;
    725 	assert(sizeof(sp->stream.avail_in)==4);  /* if this assert gets raised,
    726 	    we need to simplify this code to reflect a ZLib that is likely updated
    727 	    to deal with 8byte memory sizes, though this code will respond
    728 	    apropriately even before we simplify it */
    729 	sp->stream.avail_in = (uInt) tif->tif_rawcc;
    730 	if ((tmsize_t)sp->stream.avail_in != tif->tif_rawcc)
    731 	{
    732 		TIFFErrorExt(tif->tif_clientdata, module, "ZLib cannot deal with buffers this size");
    733 		return (0);
    734 	}
    735 	return (inflateReset(&sp->stream) == Z_OK);
    736 }
    737 
    738 static int
    739 PixarLogDecode(TIFF* tif, uint8* op, tmsize_t occ, uint16 s)
    740 {
    741 	static const char module[] = "PixarLogDecode";
    742 	TIFFDirectory *td = &tif->tif_dir;
    743 	PixarLogState* sp = DecoderState(tif);
    744 	tmsize_t i;
    745 	tmsize_t nsamples;
    746 	int llen;
    747 	uint16 *up;
    748 
    749 	switch (sp->user_datafmt) {
    750 	case PIXARLOGDATAFMT_FLOAT:
    751 		nsamples = occ / sizeof(float);	/* XXX float == 32 bits */
    752 		break;
    753 	case PIXARLOGDATAFMT_16BIT:
    754 	case PIXARLOGDATAFMT_12BITPICIO:
    755 	case PIXARLOGDATAFMT_11BITLOG:
    756 		nsamples = occ / sizeof(uint16); /* XXX uint16 == 16 bits */
    757 		break;
    758 	case PIXARLOGDATAFMT_8BIT:
    759 	case PIXARLOGDATAFMT_8BITABGR:
    760 		nsamples = occ;
    761 		break;
    762 	default:
    763 		TIFFErrorExt(tif->tif_clientdata, module,
    764 			"%d bit input not supported in PixarLog",
    765 			td->td_bitspersample);
    766 		return 0;
    767 	}
    768 
    769 	llen = sp->stride * td->td_imagewidth;
    770 
    771 	(void) s;
    772 	assert(sp != NULL);
    773 	sp->stream.next_out = (unsigned char *) sp->tbuf;
    774 	assert(sizeof(sp->stream.avail_out)==4);  /* if this assert gets raised,
    775 	    we need to simplify this code to reflect a ZLib that is likely updated
    776 	    to deal with 8byte memory sizes, though this code will respond
    777 	    apropriately even before we simplify it */
    778 	sp->stream.avail_out = (uInt) (nsamples * sizeof(uint16));
    779 	if (sp->stream.avail_out != nsamples * sizeof(uint16))
    780 	{
    781 		TIFFErrorExt(tif->tif_clientdata, module, "ZLib cannot deal with buffers this size");
    782 		return (0);
    783 	}
    784 	do {
    785 		int state = inflate(&sp->stream, Z_PARTIAL_FLUSH);
    786 		if (state == Z_STREAM_END) {
    787 			break;			/* XXX */
    788 		}
    789 		if (state == Z_DATA_ERROR) {
    790 			TIFFErrorExt(tif->tif_clientdata, module,
    791 			    "Decoding error at scanline %lu, %s",
    792 			    (unsigned long) tif->tif_row, sp->stream.msg);
    793 			if (inflateSync(&sp->stream) != Z_OK)
    794 				return (0);
    795 			continue;
    796 		}
    797 		if (state != Z_OK) {
    798 			TIFFErrorExt(tif->tif_clientdata, module, "ZLib error: %s",
    799 			    sp->stream.msg);
    800 			return (0);
    801 		}
    802 	} while (sp->stream.avail_out > 0);
    803 
    804 	/* hopefully, we got all the bytes we needed */
    805 	if (sp->stream.avail_out != 0) {
    806 		TIFFErrorExt(tif->tif_clientdata, module,
    807 		    "Not enough data at scanline %lu (short " TIFF_UINT64_FORMAT " bytes)",
    808 		    (unsigned long) tif->tif_row, (TIFF_UINT64_T) sp->stream.avail_out);
    809 		return (0);
    810 	}
    811 
    812 	up = sp->tbuf;
    813 	/* Swap bytes in the data if from a different endian machine. */
    814 	if (tif->tif_flags & TIFF_SWAB)
    815 		TIFFSwabArrayOfShort(up, nsamples);
    816 
    817 	/*
    818 	 * if llen is not an exact multiple of nsamples, the decode operation
    819 	 * may overflow the output buffer, so truncate it enough to prevent
    820 	 * that but still salvage as much data as possible.
    821 	 */
    822 	if (nsamples % llen) {
    823 		TIFFWarningExt(tif->tif_clientdata, module,
    824 			"stride %lu is not a multiple of sample count, "
    825 			"%lu, data truncated.", (unsigned long) llen, (unsigned long) nsamples);
    826 		nsamples -= nsamples % llen;
    827 	}
    828 
    829 	for (i = 0; i < nsamples; i += llen, up += llen) {
    830 		switch (sp->user_datafmt)  {
    831 		case PIXARLOGDATAFMT_FLOAT:
    832 			horizontalAccumulateF(up, llen, sp->stride,
    833 					(float *)op, sp->ToLinearF);
    834 			op += llen * sizeof(float);
    835 			break;
    836 		case PIXARLOGDATAFMT_16BIT:
    837 			horizontalAccumulate16(up, llen, sp->stride,
    838 					(uint16 *)op, sp->ToLinear16);
    839 			op += llen * sizeof(uint16);
    840 			break;
    841 		case PIXARLOGDATAFMT_12BITPICIO:
    842 			horizontalAccumulate12(up, llen, sp->stride,
    843 					(int16 *)op, sp->ToLinearF);
    844 			op += llen * sizeof(int16);
    845 			break;
    846 		case PIXARLOGDATAFMT_11BITLOG:
    847 			horizontalAccumulate11(up, llen, sp->stride,
    848 					(uint16 *)op);
    849 			op += llen * sizeof(uint16);
    850 			break;
    851 		case PIXARLOGDATAFMT_8BIT:
    852 			horizontalAccumulate8(up, llen, sp->stride,
    853 					(unsigned char *)op, sp->ToLinear8);
    854 			op += llen * sizeof(unsigned char);
    855 			break;
    856 		case PIXARLOGDATAFMT_8BITABGR:
    857 			horizontalAccumulate8abgr(up, llen, sp->stride,
    858 					(unsigned char *)op, sp->ToLinear8);
    859 			op += llen * sizeof(unsigned char);
    860 			break;
    861 		default:
    862 			TIFFErrorExt(tif->tif_clientdata, module,
    863 				  "Unsupported bits/sample: %d",
    864 				  td->td_bitspersample);
    865 			return (0);
    866 		}
    867 	}
    868 
    869 	return (1);
    870 }
    871 
    872 static int
    873 PixarLogSetupEncode(TIFF* tif)
    874 {
    875 	static const char module[] = "PixarLogSetupEncode";
    876 	TIFFDirectory *td = &tif->tif_dir;
    877 	PixarLogState* sp = EncoderState(tif);
    878 	tmsize_t tbuf_size;
    879 
    880 	assert(sp != NULL);
    881 
    882 	/* for some reason, we can't do this in TIFFInitPixarLog */
    883 
    884 	sp->stride = (td->td_planarconfig == PLANARCONFIG_CONTIG ?
    885 	    td->td_samplesperpixel : 1);
    886 	tbuf_size = multiply_ms(multiply_ms(multiply_ms(sp->stride, td->td_imagewidth),
    887 				      td->td_rowsperstrip), sizeof(uint16));
    888 	if (tbuf_size == 0)
    889 		return (0);  /* TODO: this is an error return without error report through TIFFErrorExt */
    890 	sp->tbuf = (uint16 *) _TIFFmalloc(tbuf_size);
    891 	if (sp->tbuf == NULL)
    892 		return (0);
    893 	if (sp->user_datafmt == PIXARLOGDATAFMT_UNKNOWN)
    894 		sp->user_datafmt = PixarLogGuessDataFmt(td);
    895 	if (sp->user_datafmt == PIXARLOGDATAFMT_UNKNOWN) {
    896 		TIFFErrorExt(tif->tif_clientdata, module, "PixarLog compression can't handle %d bit linear encodings", td->td_bitspersample);
    897 		return (0);
    898 	}
    899 
    900 	if (deflateInit(&sp->stream, sp->quality) != Z_OK) {
    901 		TIFFErrorExt(tif->tif_clientdata, module, "%s", sp->stream.msg);
    902 		return (0);
    903 	} else {
    904 		sp->state |= PLSTATE_INIT;
    905 		return (1);
    906 	}
    907 }
    908 
    909 /*
    910  * Reset encoding state at the start of a strip.
    911  */
    912 static int
    913 PixarLogPreEncode(TIFF* tif, uint16 s)
    914 {
    915 	static const char module[] = "PixarLogPreEncode";
    916 	PixarLogState *sp = EncoderState(tif);
    917 
    918 	(void) s;
    919 	assert(sp != NULL);
    920 	sp->stream.next_out = tif->tif_rawdata;
    921 	assert(sizeof(sp->stream.avail_out)==4);  /* if this assert gets raised,
    922 	    we need to simplify this code to reflect a ZLib that is likely updated
    923 	    to deal with 8byte memory sizes, though this code will respond
    924 	    apropriately even before we simplify it */
    925 	sp->stream.avail_out = tif->tif_rawdatasize;
    926 	if ((tmsize_t)sp->stream.avail_out != tif->tif_rawdatasize)
    927 	{
    928 		TIFFErrorExt(tif->tif_clientdata, module, "ZLib cannot deal with buffers this size");
    929 		return (0);
    930 	}
    931 	return (deflateReset(&sp->stream) == Z_OK);
    932 }
    933 
    934 static void
    935 horizontalDifferenceF(float *ip, int n, int stride, uint16 *wp, uint16 *FromLT2)
    936 {
    937     int32 r1, g1, b1, a1, r2, g2, b2, a2, mask;
    938     float fltsize = Fltsize;
    939 
    940 #define  CLAMP(v) ( (v<(float)0.)   ? 0				\
    941 		  : (v<(float)2.)   ? FromLT2[(int)(v*fltsize)]	\
    942 		  : (v>(float)24.2) ? 2047			\
    943 		  : LogK1*log(v*LogK2) + 0.5 )
    944 
    945     mask = CODE_MASK;
    946     if (n >= stride) {
    947 	if (stride == 3) {
    948 	    r2 = wp[0] = (uint16) CLAMP(ip[0]);
    949 	    g2 = wp[1] = (uint16) CLAMP(ip[1]);
    950 	    b2 = wp[2] = (uint16) CLAMP(ip[2]);
    951 	    n -= 3;
    952 	    while (n > 0) {
    953 		n -= 3;
    954 		wp += 3;
    955 		ip += 3;
    956 		r1 = (int32) CLAMP(ip[0]); wp[0] = (r1-r2) & mask; r2 = r1;
    957 		g1 = (int32) CLAMP(ip[1]); wp[1] = (g1-g2) & mask; g2 = g1;
    958 		b1 = (int32) CLAMP(ip[2]); wp[2] = (b1-b2) & mask; b2 = b1;
    959 	    }
    960 	} else if (stride == 4) {
    961 	    r2 = wp[0] = (uint16) CLAMP(ip[0]);
    962 	    g2 = wp[1] = (uint16) CLAMP(ip[1]);
    963 	    b2 = wp[2] = (uint16) CLAMP(ip[2]);
    964 	    a2 = wp[3] = (uint16) CLAMP(ip[3]);
    965 	    n -= 4;
    966 	    while (n > 0) {
    967 		n -= 4;
    968 		wp += 4;
    969 		ip += 4;
    970 		r1 = (int32) CLAMP(ip[0]); wp[0] = (r1-r2) & mask; r2 = r1;
    971 		g1 = (int32) CLAMP(ip[1]); wp[1] = (g1-g2) & mask; g2 = g1;
    972 		b1 = (int32) CLAMP(ip[2]); wp[2] = (b1-b2) & mask; b2 = b1;
    973 		a1 = (int32) CLAMP(ip[3]); wp[3] = (a1-a2) & mask; a2 = a1;
    974 	    }
    975 	} else {
    976 	    ip += n - 1;	/* point to last one */
    977 	    wp += n - 1;	/* point to last one */
    978 	    n -= stride;
    979 	    while (n > 0) {
    980 		REPEAT(stride, wp[0] = (uint16) CLAMP(ip[0]);
    981 				wp[stride] -= wp[0];
    982 				wp[stride] &= mask;
    983 				wp--; ip--)
    984 		n -= stride;
    985 	    }
    986 	    REPEAT(stride, wp[0] = (uint16) CLAMP(ip[0]); wp--; ip--)
    987 	}
    988     }
    989 }
    990 
    991 static void
    992 horizontalDifference16(unsigned short *ip, int n, int stride,
    993 	unsigned short *wp, uint16 *From14)
    994 {
    995     register int  r1, g1, b1, a1, r2, g2, b2, a2, mask;
    996 
    997 /* assumption is unsigned pixel values */
    998 #undef   CLAMP
    999 #define  CLAMP(v) From14[(v) >> 2]
   1000 
   1001     mask = CODE_MASK;
   1002     if (n >= stride) {
   1003 	if (stride == 3) {
   1004 	    r2 = wp[0] = CLAMP(ip[0]);  g2 = wp[1] = CLAMP(ip[1]);
   1005 	    b2 = wp[2] = CLAMP(ip[2]);
   1006 	    n -= 3;
   1007 	    while (n > 0) {
   1008 		n -= 3;
   1009 		wp += 3;
   1010 		ip += 3;
   1011 		r1 = CLAMP(ip[0]); wp[0] = (r1-r2) & mask; r2 = r1;
   1012 		g1 = CLAMP(ip[1]); wp[1] = (g1-g2) & mask; g2 = g1;
   1013 		b1 = CLAMP(ip[2]); wp[2] = (b1-b2) & mask; b2 = b1;
   1014 	    }
   1015 	} else if (stride == 4) {
   1016 	    r2 = wp[0] = CLAMP(ip[0]);  g2 = wp[1] = CLAMP(ip[1]);
   1017 	    b2 = wp[2] = CLAMP(ip[2]);  a2 = wp[3] = CLAMP(ip[3]);
   1018 	    n -= 4;
   1019 	    while (n > 0) {
   1020 		n -= 4;
   1021 		wp += 4;
   1022 		ip += 4;
   1023 		r1 = CLAMP(ip[0]); wp[0] = (r1-r2) & mask; r2 = r1;
   1024 		g1 = CLAMP(ip[1]); wp[1] = (g1-g2) & mask; g2 = g1;
   1025 		b1 = CLAMP(ip[2]); wp[2] = (b1-b2) & mask; b2 = b1;
   1026 		a1 = CLAMP(ip[3]); wp[3] = (a1-a2) & mask; a2 = a1;
   1027 	    }
   1028 	} else {
   1029 	    ip += n - 1;	/* point to last one */
   1030 	    wp += n - 1;	/* point to last one */
   1031 	    n -= stride;
   1032 	    while (n > 0) {
   1033 		REPEAT(stride, wp[0] = CLAMP(ip[0]);
   1034 				wp[stride] -= wp[0];
   1035 				wp[stride] &= mask;
   1036 				wp--; ip--)
   1037 		n -= stride;
   1038 	    }
   1039 	    REPEAT(stride, wp[0] = CLAMP(ip[0]); wp--; ip--)
   1040 	}
   1041     }
   1042 }
   1043 
   1044 
   1045 static void
   1046 horizontalDifference8(unsigned char *ip, int n, int stride,
   1047 	unsigned short *wp, uint16 *From8)
   1048 {
   1049     register int  r1, g1, b1, a1, r2, g2, b2, a2, mask;
   1050 
   1051 #undef	 CLAMP
   1052 #define  CLAMP(v) (From8[(v)])
   1053 
   1054     mask = CODE_MASK;
   1055     if (n >= stride) {
   1056 	if (stride == 3) {
   1057 	    r2 = wp[0] = CLAMP(ip[0]);  g2 = wp[1] = CLAMP(ip[1]);
   1058 	    b2 = wp[2] = CLAMP(ip[2]);
   1059 	    n -= 3;
   1060 	    while (n > 0) {
   1061 		n -= 3;
   1062 		r1 = CLAMP(ip[3]); wp[3] = (r1-r2) & mask; r2 = r1;
   1063 		g1 = CLAMP(ip[4]); wp[4] = (g1-g2) & mask; g2 = g1;
   1064 		b1 = CLAMP(ip[5]); wp[5] = (b1-b2) & mask; b2 = b1;
   1065 		wp += 3;
   1066 		ip += 3;
   1067 	    }
   1068 	} else if (stride == 4) {
   1069 	    r2 = wp[0] = CLAMP(ip[0]);  g2 = wp[1] = CLAMP(ip[1]);
   1070 	    b2 = wp[2] = CLAMP(ip[2]);  a2 = wp[3] = CLAMP(ip[3]);
   1071 	    n -= 4;
   1072 	    while (n > 0) {
   1073 		n -= 4;
   1074 		r1 = CLAMP(ip[4]); wp[4] = (r1-r2) & mask; r2 = r1;
   1075 		g1 = CLAMP(ip[5]); wp[5] = (g1-g2) & mask; g2 = g1;
   1076 		b1 = CLAMP(ip[6]); wp[6] = (b1-b2) & mask; b2 = b1;
   1077 		a1 = CLAMP(ip[7]); wp[7] = (a1-a2) & mask; a2 = a1;
   1078 		wp += 4;
   1079 		ip += 4;
   1080 	    }
   1081 	} else {
   1082 	    wp += n + stride - 1;	/* point to last one */
   1083 	    ip += n + stride - 1;	/* point to last one */
   1084 	    n -= stride;
   1085 	    while (n > 0) {
   1086 		REPEAT(stride, wp[0] = CLAMP(ip[0]);
   1087 				wp[stride] -= wp[0];
   1088 				wp[stride] &= mask;
   1089 				wp--; ip--)
   1090 		n -= stride;
   1091 	    }
   1092 	    REPEAT(stride, wp[0] = CLAMP(ip[0]); wp--; ip--)
   1093 	}
   1094     }
   1095 }
   1096 
   1097 /*
   1098  * Encode a chunk of pixels.
   1099  */
   1100 static int
   1101 PixarLogEncode(TIFF* tif, uint8* bp, tmsize_t cc, uint16 s)
   1102 {
   1103 	static const char module[] = "PixarLogEncode";
   1104 	TIFFDirectory *td = &tif->tif_dir;
   1105 	PixarLogState *sp = EncoderState(tif);
   1106 	tmsize_t i;
   1107 	tmsize_t n;
   1108 	int llen;
   1109 	unsigned short * up;
   1110 
   1111 	(void) s;
   1112 
   1113 	switch (sp->user_datafmt) {
   1114 	case PIXARLOGDATAFMT_FLOAT:
   1115 		n = cc / sizeof(float);		/* XXX float == 32 bits */
   1116 		break;
   1117 	case PIXARLOGDATAFMT_16BIT:
   1118 	case PIXARLOGDATAFMT_12BITPICIO:
   1119 	case PIXARLOGDATAFMT_11BITLOG:
   1120 		n = cc / sizeof(uint16);	/* XXX uint16 == 16 bits */
   1121 		break;
   1122 	case PIXARLOGDATAFMT_8BIT:
   1123 	case PIXARLOGDATAFMT_8BITABGR:
   1124 		n = cc;
   1125 		break;
   1126 	default:
   1127 		TIFFErrorExt(tif->tif_clientdata, module,
   1128 			"%d bit input not supported in PixarLog",
   1129 			td->td_bitspersample);
   1130 		return 0;
   1131 	}
   1132 
   1133 	llen = sp->stride * td->td_imagewidth;
   1134 
   1135 	for (i = 0, up = sp->tbuf; i < n; i += llen, up += llen) {
   1136 		switch (sp->user_datafmt)  {
   1137 		case PIXARLOGDATAFMT_FLOAT:
   1138 			horizontalDifferenceF((float *)bp, llen,
   1139 				sp->stride, up, sp->FromLT2);
   1140 			bp += llen * sizeof(float);
   1141 			break;
   1142 		case PIXARLOGDATAFMT_16BIT:
   1143 			horizontalDifference16((uint16 *)bp, llen,
   1144 				sp->stride, up, sp->From14);
   1145 			bp += llen * sizeof(uint16);
   1146 			break;
   1147 		case PIXARLOGDATAFMT_8BIT:
   1148 			horizontalDifference8((unsigned char *)bp, llen,
   1149 				sp->stride, up, sp->From8);
   1150 			bp += llen * sizeof(unsigned char);
   1151 			break;
   1152 		default:
   1153 			TIFFErrorExt(tif->tif_clientdata, module,
   1154 				"%d bit input not supported in PixarLog",
   1155 				td->td_bitspersample);
   1156 			return 0;
   1157 		}
   1158 	}
   1159 
   1160 	sp->stream.next_in = (unsigned char *) sp->tbuf;
   1161 	assert(sizeof(sp->stream.avail_in)==4);  /* if this assert gets raised,
   1162 	    we need to simplify this code to reflect a ZLib that is likely updated
   1163 	    to deal with 8byte memory sizes, though this code will respond
   1164 	    apropriately even before we simplify it */
   1165 	sp->stream.avail_in = (uInt) (n * sizeof(uint16));
   1166 	if ((sp->stream.avail_in / sizeof(uint16)) != (uInt) n)
   1167 	{
   1168 		TIFFErrorExt(tif->tif_clientdata, module,
   1169 			     "ZLib cannot deal with buffers this size");
   1170 		return (0);
   1171 	}
   1172 
   1173 	do {
   1174 		if (deflate(&sp->stream, Z_NO_FLUSH) != Z_OK) {
   1175 			TIFFErrorExt(tif->tif_clientdata, module, "Encoder error: %s",
   1176 			    sp->stream.msg);
   1177 			return (0);
   1178 		}
   1179 		if (sp->stream.avail_out == 0) {
   1180 			tif->tif_rawcc = tif->tif_rawdatasize;
   1181 			TIFFFlushData1(tif);
   1182 			sp->stream.next_out = tif->tif_rawdata;
   1183 			sp->stream.avail_out = (uInt) tif->tif_rawdatasize;  /* this is a safe typecast, as check is made already in PixarLogPreEncode */
   1184 		}
   1185 	} while (sp->stream.avail_in > 0);
   1186 	return (1);
   1187 }
   1188 
   1189 /*
   1190  * Finish off an encoded strip by flushing the last
   1191  * string and tacking on an End Of Information code.
   1192  */
   1193 
   1194 static int
   1195 PixarLogPostEncode(TIFF* tif)
   1196 {
   1197 	static const char module[] = "PixarLogPostEncode";
   1198 	PixarLogState *sp = EncoderState(tif);
   1199 	int state;
   1200 
   1201 	sp->stream.avail_in = 0;
   1202 
   1203 	do {
   1204 		state = deflate(&sp->stream, Z_FINISH);
   1205 		switch (state) {
   1206 		case Z_STREAM_END:
   1207 		case Z_OK:
   1208 		    if ((tmsize_t)sp->stream.avail_out != tif->tif_rawdatasize) {
   1209 			    tif->tif_rawcc =
   1210 				tif->tif_rawdatasize - sp->stream.avail_out;
   1211 			    TIFFFlushData1(tif);
   1212 			    sp->stream.next_out = tif->tif_rawdata;
   1213 			    sp->stream.avail_out = (uInt) tif->tif_rawdatasize;  /* this is a safe typecast, as check is made already in PixarLogPreEncode */
   1214 		    }
   1215 		    break;
   1216 		default:
   1217 			TIFFErrorExt(tif->tif_clientdata, module, "ZLib error: %s",
   1218 			sp->stream.msg);
   1219 		    return (0);
   1220 		}
   1221 	} while (state != Z_STREAM_END);
   1222 	return (1);
   1223 }
   1224 
   1225 static void
   1226 PixarLogClose(TIFF* tif)
   1227 {
   1228 	TIFFDirectory *td = &tif->tif_dir;
   1229 
   1230 	/* In a really sneaky (and really incorrect, and untruthfull, and
   1231 	 * troublesome, and error-prone) maneuver that completely goes against
   1232 	 * the spirit of TIFF, and breaks TIFF, on close, we covertly
   1233 	 * modify both bitspersample and sampleformat in the directory to
   1234 	 * indicate 8-bit linear.  This way, the decode "just works" even for
   1235 	 * readers that don't know about PixarLog, or how to set
   1236 	 * the PIXARLOGDATFMT pseudo-tag.
   1237 	 */
   1238 	td->td_bitspersample = 8;
   1239 	td->td_sampleformat = SAMPLEFORMAT_UINT;
   1240 }
   1241 
   1242 static void
   1243 PixarLogCleanup(TIFF* tif)
   1244 {
   1245 	PixarLogState* sp = (PixarLogState*) tif->tif_data;
   1246 
   1247 	assert(sp != 0);
   1248 
   1249 	(void)TIFFPredictorCleanup(tif);
   1250 
   1251 	tif->tif_tagmethods.vgetfield = sp->vgetparent;
   1252 	tif->tif_tagmethods.vsetfield = sp->vsetparent;
   1253 
   1254 	if (sp->FromLT2) _TIFFfree(sp->FromLT2);
   1255 	if (sp->From14) _TIFFfree(sp->From14);
   1256 	if (sp->From8) _TIFFfree(sp->From8);
   1257 	if (sp->ToLinearF) _TIFFfree(sp->ToLinearF);
   1258 	if (sp->ToLinear16) _TIFFfree(sp->ToLinear16);
   1259 	if (sp->ToLinear8) _TIFFfree(sp->ToLinear8);
   1260 	if (sp->state&PLSTATE_INIT) {
   1261 		if (tif->tif_mode == O_RDONLY)
   1262 			inflateEnd(&sp->stream);
   1263 		else
   1264 			deflateEnd(&sp->stream);
   1265 	}
   1266 	if (sp->tbuf)
   1267 		_TIFFfree(sp->tbuf);
   1268 	_TIFFfree(sp);
   1269 	tif->tif_data = NULL;
   1270 
   1271 	_TIFFSetDefaultCompressionState(tif);
   1272 }
   1273 
   1274 static int
   1275 PixarLogVSetField(TIFF* tif, uint32 tag, va_list ap)
   1276 {
   1277     static const char module[] = "PixarLogVSetField";
   1278     PixarLogState *sp = (PixarLogState *)tif->tif_data;
   1279     int result;
   1280 
   1281     switch (tag) {
   1282      case TIFFTAG_PIXARLOGQUALITY:
   1283 		sp->quality = (int) va_arg(ap, int);
   1284 		if (tif->tif_mode != O_RDONLY && (sp->state&PLSTATE_INIT)) {
   1285 			if (deflateParams(&sp->stream,
   1286 			    sp->quality, Z_DEFAULT_STRATEGY) != Z_OK) {
   1287 				TIFFErrorExt(tif->tif_clientdata, module, "ZLib error: %s",
   1288 					sp->stream.msg);
   1289 				return (0);
   1290 			}
   1291 		}
   1292 		return (1);
   1293      case TIFFTAG_PIXARLOGDATAFMT:
   1294 	sp->user_datafmt = (int) va_arg(ap, int);
   1295 	/* Tweak the TIFF header so that the rest of libtiff knows what
   1296 	 * size of data will be passed between app and library, and
   1297 	 * assume that the app knows what it is doing and is not
   1298 	 * confused by these header manipulations...
   1299 	 */
   1300 	switch (sp->user_datafmt) {
   1301 	 case PIXARLOGDATAFMT_8BIT:
   1302 	 case PIXARLOGDATAFMT_8BITABGR:
   1303 	    TIFFSetField(tif, TIFFTAG_BITSPERSAMPLE, 8);
   1304 	    TIFFSetField(tif, TIFFTAG_SAMPLEFORMAT, SAMPLEFORMAT_UINT);
   1305 	    break;
   1306 	 case PIXARLOGDATAFMT_11BITLOG:
   1307 	    TIFFSetField(tif, TIFFTAG_BITSPERSAMPLE, 16);
   1308 	    TIFFSetField(tif, TIFFTAG_SAMPLEFORMAT, SAMPLEFORMAT_UINT);
   1309 	    break;
   1310 	 case PIXARLOGDATAFMT_12BITPICIO:
   1311 	    TIFFSetField(tif, TIFFTAG_BITSPERSAMPLE, 16);
   1312 	    TIFFSetField(tif, TIFFTAG_SAMPLEFORMAT, SAMPLEFORMAT_INT);
   1313 	    break;
   1314 	 case PIXARLOGDATAFMT_16BIT:
   1315 	    TIFFSetField(tif, TIFFTAG_BITSPERSAMPLE, 16);
   1316 	    TIFFSetField(tif, TIFFTAG_SAMPLEFORMAT, SAMPLEFORMAT_UINT);
   1317 	    break;
   1318 	 case PIXARLOGDATAFMT_FLOAT:
   1319 	    TIFFSetField(tif, TIFFTAG_BITSPERSAMPLE, 32);
   1320 	    TIFFSetField(tif, TIFFTAG_SAMPLEFORMAT, SAMPLEFORMAT_IEEEFP);
   1321 	    break;
   1322 	}
   1323 	/*
   1324 	 * Must recalculate sizes should bits/sample change.
   1325 	 */
   1326 	tif->tif_tilesize = isTiled(tif) ? TIFFTileSize(tif) : (tmsize_t)(-1);
   1327 	tif->tif_scanlinesize = TIFFScanlineSize(tif);
   1328 	result = 1;		/* NB: pseudo tag */
   1329 	break;
   1330      default:
   1331 	result = (*sp->vsetparent)(tif, tag, ap);
   1332     }
   1333     return (result);
   1334 }
   1335 
   1336 static int
   1337 PixarLogVGetField(TIFF* tif, uint32 tag, va_list ap)
   1338 {
   1339     PixarLogState *sp = (PixarLogState *)tif->tif_data;
   1340 
   1341     switch (tag) {
   1342      case TIFFTAG_PIXARLOGQUALITY:
   1343 	*va_arg(ap, int*) = sp->quality;
   1344 	break;
   1345      case TIFFTAG_PIXARLOGDATAFMT:
   1346 	*va_arg(ap, int*) = sp->user_datafmt;
   1347 	break;
   1348      default:
   1349 	return (*sp->vgetparent)(tif, tag, ap);
   1350     }
   1351     return (1);
   1352 }
   1353 
   1354 static const TIFFField pixarlogFields[] = {
   1355     {TIFFTAG_PIXARLOGDATAFMT, 0, 0, TIFF_ANY, 0, TIFF_SETGET_INT, TIFF_SETGET_UNDEFINED, FIELD_PSEUDO, FALSE, FALSE, "", NULL},
   1356     {TIFFTAG_PIXARLOGQUALITY, 0, 0, TIFF_ANY, 0, TIFF_SETGET_INT, TIFF_SETGET_UNDEFINED, FIELD_PSEUDO, FALSE, FALSE, "", NULL}
   1357 };
   1358 
   1359 int
   1360 TIFFInitPixarLog(TIFF* tif, int scheme)
   1361 {
   1362 	static const char module[] = "TIFFInitPixarLog";
   1363 
   1364 	PixarLogState* sp;
   1365 
   1366 	assert(scheme == COMPRESSION_PIXARLOG);
   1367 
   1368 	/*
   1369 	 * Merge codec-specific tag information.
   1370 	 */
   1371 	if (!_TIFFMergeFields(tif, pixarlogFields,
   1372 			      TIFFArrayCount(pixarlogFields))) {
   1373 		TIFFErrorExt(tif->tif_clientdata, module,
   1374 			     "Merging PixarLog codec-specific tags failed");
   1375 		return 0;
   1376 	}
   1377 
   1378 	/*
   1379 	 * Allocate state block so tag methods have storage to record values.
   1380 	 */
   1381 	tif->tif_data = (uint8*) _TIFFmalloc(sizeof (PixarLogState));
   1382 	if (tif->tif_data == NULL)
   1383 		goto bad;
   1384 	sp = (PixarLogState*) tif->tif_data;
   1385 	_TIFFmemset(sp, 0, sizeof (*sp));
   1386 	sp->stream.data_type = Z_BINARY;
   1387 	sp->user_datafmt = PIXARLOGDATAFMT_UNKNOWN;
   1388 
   1389 	/*
   1390 	 * Install codec methods.
   1391 	 */
   1392 	tif->tif_fixuptags = PixarLogFixupTags;
   1393 	tif->tif_setupdecode = PixarLogSetupDecode;
   1394 	tif->tif_predecode = PixarLogPreDecode;
   1395 	tif->tif_decoderow = PixarLogDecode;
   1396 	tif->tif_decodestrip = PixarLogDecode;
   1397 	tif->tif_decodetile = PixarLogDecode;
   1398 	tif->tif_setupencode = PixarLogSetupEncode;
   1399 	tif->tif_preencode = PixarLogPreEncode;
   1400 	tif->tif_postencode = PixarLogPostEncode;
   1401 	tif->tif_encoderow = PixarLogEncode;
   1402 	tif->tif_encodestrip = PixarLogEncode;
   1403 	tif->tif_encodetile = PixarLogEncode;
   1404 	tif->tif_close = PixarLogClose;
   1405 	tif->tif_cleanup = PixarLogCleanup;
   1406 
   1407 	/* Override SetField so we can handle our private pseudo-tag */
   1408 	sp->vgetparent = tif->tif_tagmethods.vgetfield;
   1409 	tif->tif_tagmethods.vgetfield = PixarLogVGetField;   /* hook for codec tags */
   1410 	sp->vsetparent = tif->tif_tagmethods.vsetfield;
   1411 	tif->tif_tagmethods.vsetfield = PixarLogVSetField;   /* hook for codec tags */
   1412 
   1413 	/* Default values for codec-specific fields */
   1414 	sp->quality = Z_DEFAULT_COMPRESSION; /* default comp. level */
   1415 	sp->state = 0;
   1416 
   1417 	/* we don't wish to use the predictor,
   1418 	 * the default is none, which predictor value 1
   1419 	 */
   1420 	(void) TIFFPredictorInit(tif);
   1421 
   1422 	/*
   1423 	 * build the companding tables
   1424 	 */
   1425 	PixarLogMakeTables(sp);
   1426 
   1427 	return (1);
   1428 bad:
   1429 	TIFFErrorExt(tif->tif_clientdata, module,
   1430 		     "No space for PixarLog state block");
   1431 	return (0);
   1432 }
   1433 #endif /* PIXARLOG_SUPPORT */
   1434 
   1435 /* vim: set ts=8 sts=8 sw=8 noet: */
   1436 /*
   1437  * Local Variables:
   1438  * mode: c
   1439  * c-basic-offset: 8
   1440  * fill-column: 78
   1441  * End:
   1442  */
   1443