tif_luv.c
上传用户:looem2003
上传日期:2014-07-20
资源大小:13733k
文件大小:40k
- /* $Id: tif_luv.c,v 1.17 2006/03/16 12:38:24 dron 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 (0);
- }
- /*
- * 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 (0);
- }
- /*
- * 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 (0);
- }
- /*
- * 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) == 0)
- 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) == 0)
- 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) {
- u = U_NEU; v = V_NEU;
- }
- s = 1./(6.*u - 16.*v + 12.);
- x = 9.*u * s;
- y = 4.*v * s;
- /* convert to XYZ */
- XYZ[0] = (float)(x/y * L);
- XYZ[1] = (float)L;
- XYZ[2] = (float)((1.-x-y)/y * L);
- }
- #if !LOGLUV_PUBLIC
- static
- #endif
- uint32
- LogLuv24fromXYZ(float XYZ[3], int em)
- {
- int Le, Ce;
- double u, v, s;
- /* encode luminance */
- Le = LogL10fromY(XYZ[1], em);
- /* encode color */
- s = XYZ[0] + 15.*XYZ[1] + 3.*XYZ[2];
- if (!Le || s <= 0.) {
- u = U_NEU;
- v = V_NEU;
- } else {
- u = 4.*XYZ[0] / s;
- v = 9.*XYZ[1] / s;
- }
- Ce = uv_encode(u, v, em);
- if (Ce < 0) /* never happens */
- Ce = uv_encode(U_NEU, V_NEU, SGILOGENCODE_NODITHER);
- /* combine encodings */
- return (Le << 14 | Ce);
- }
- static void
- Luv24toXYZ(LogLuvState* sp, tidata_t op, int n)
- {
- uint32* luv = (uint32*) sp->tbuf;
- float* xyz = (float*) op;
- while (n-- > 0) {
- LogLuv24toXYZ(*luv, xyz);
- xyz += 3;
- luv++;
- }
- }
- static void
- Luv24toLuv48(LogLuvState* sp, tidata_t op, int n)
- {
- uint32* luv = (uint32*) sp->tbuf;
- int16* luv3 = (int16*) op;
- while (n-- > 0) {
- double u, v;
- *luv3++ = (int16)((*luv >> 12 & 0xffd) + 13314);
- if (uv_decode(&u, &v, *luv&0x3fff) < 0) {
- u = U_NEU;
- v = V_NEU;
- }
- *luv3++ = (int16)(u * (1L<<15));
- *luv3++ = (int16)(v * (1L<<15));
- luv++;
- }
- }
- static void
- Luv24toRGB(LogLuvState* sp, tidata_t op, int n)
- {
- uint32* luv = (uint32*) sp->tbuf;
- uint8* rgb = (uint8*) op;
- while (n-- > 0) {
- float xyz[3];
- LogLuv24toXYZ(*luv++, xyz);
- XYZtoRGB24(xyz, rgb);
- rgb += 3;
- }
- }
- static void
- Luv24fromXYZ(LogLuvState* sp, tidata_t op, int n)
- {
- uint32* luv = (uint32*) sp->tbuf;
- float* xyz = (float*) op;
- while (n-- > 0) {
- *luv++ = LogLuv24fromXYZ(xyz, sp->encode_meth);
- xyz += 3;
- }
- }
- static void
- Luv24fromLuv48(LogLuvState* sp, tidata_t op, int n)
- {
- uint32* luv = (uint32*) sp->tbuf;
- int16* luv3 = (int16*) op;
- while (n-- > 0) {
- int Le, Ce;
- if (luv3[0] <= 0)
- Le = 0;
- else if (luv3[0] >= (1<<12)+3314)
- Le = (1<<10) - 1;
- else if (sp->encode_meth == SGILOGENCODE_NODITHER)
- Le = (luv3[0]-3314) >> 2;
- else
- Le = itrunc(.25*(luv3[0]-3314.), sp->encode_meth);
- Ce = uv_encode((luv3[1]+.5)/(1<<15), (luv3[2]+.5)/(1<<15),
- sp->encode_meth);
- if (Ce < 0) /* never happens */
- Ce = uv_encode(U_NEU, V_NEU, SGILOGENCODE_NODITHER);
- *luv++ = (uint32)Le << 14 | Ce;
- luv3 += 3;
- }
- }
- #if !LOGLUV_PUBLIC
- static
- #endif
- void
- LogLuv32toXYZ(uint32 p, float XYZ[3])
- {
- double L, u, v, s, x, y;
- /* decode luminance */
- L = LogL16toY((int)p >> 16);
- if (L <= 0.) {
- XYZ[0] = XYZ[1] = XYZ[2] = 0.;
- return;
- }
- /* decode color */
- u = 1./UVSCALE * ((p>>8 & 0xff) + .5);
- v = 1./UVSCALE * ((p & 0xff) + .5);
- s = 1./(6.*u - 16.*v + 12.);
- x = 9.*u * s;
- y = 4.*v * s;
- /* convert to XYZ */
- XYZ[0] = (float)(x/y * L);
- XYZ[1] = (float)L;
- XYZ[2] = (float)((1.-x-y)/y * L);
- }
- #if !LOGLUV_PUBLIC
- static
- #endif
- uint32
- LogLuv32fromXYZ(float XYZ[3], int em)
- {
- unsigned int Le, ue, ve;
- double u, v, s;
- /* encode luminance */
- Le = (unsigned int)LogL16fromY(XYZ[1], em);
- /* encode color */
- s = XYZ[0] + 15.*XYZ[1] + 3.*XYZ[2];
- if (!Le || s <= 0.) {
- u = U_NEU;
- v = V_NEU;
- } else {
- u = 4.*XYZ[0] / s;
- v = 9.*XYZ[1] / s;
- }
- if (u <= 0.) ue = 0;
- else ue = itrunc(UVSCALE*u, em);
- if (ue > 255) ue = 255;
- if (v <= 0.) ve = 0;
- else ve = itrunc(UVSCALE*v, em);
- if (ve > 255) ve = 255;
- /* combine encodings */
- return (Le << 16 | ue << 8 | ve);
- }
- static void
- Luv32toXYZ(LogLuvState* sp, tidata_t op, int n)
- {
- uint32* luv = (uint32*) sp->tbuf;
- float* xyz = (float*) op;
- while (n-- > 0) {
- LogLuv32toXYZ(*luv++, xyz);
- xyz += 3;
- }
- }
- static void
- Luv32toLuv48(LogLuvState* sp, tidata_t op, int n)
- {
- uint32* luv = (uint32*) sp->tbuf;
- int16* luv3 = (int16*) op;
- while (n-- > 0) {
- double u, v;
- *luv3++ = (int16)(*luv >> 16);
- u = 1./UVSCALE * ((*luv>>8 & 0xff) + .5);
- v = 1./UVSCALE * ((*luv & 0xff) + .5);
- *luv3++ = (int16)(u * (1L<<15));
- *luv3++ = (int16)(v * (1L<<15));
- luv++;
- }
- }
- static void
- Luv32toRGB(LogLuvState* sp, tidata_t op, int n)
- {
- uint32* luv = (uint32*) sp->tbuf;
- uint8* rgb = (uint8*) op;
- while (n-- > 0) {
- float xyz[3];
- LogLuv32toXYZ(*luv++, xyz);
- XYZtoRGB24(xyz, rgb);
- rgb += 3;
- }
- }
- static void
- Luv32fromXYZ(LogLuvState* sp, tidata_t op, int n)
- {
- uint32* luv = (uint32*) sp->tbuf;
- float* xyz = (float*) op;
- while (n-- > 0) {
- *luv++ = LogLuv32fromXYZ(xyz, sp->encode_meth);
- xyz += 3;
- }
- }
- static void
- Luv32fromLuv48(LogLuvState* sp, tidata_t op, int n)
- {
- uint32* luv = (uint32*) sp->tbuf;
- int16* luv3 = (int16*) op;
- if (sp->encode_meth == SGILOGENCODE_NODITHER) {
- while (n-- > 0) {
- *luv++ = (uint32)luv3[0] << 16 |
- (luv3[1]*(uint32)(UVSCALE+.5) >> 7 & 0xff00) |
- (luv3[2]*(uint32)(UVSCALE+.5) >> 15 & 0xff);
- luv3 += 3;
- }
- return;
- }
- while (n-- > 0) {
- *luv++ = (uint32)luv3[0] << 16 |
- (itrunc(luv3[1]*(UVSCALE/(1<<15)), sp->encode_meth) << 8 & 0xff00) |
- (itrunc(luv3[2]*(UVSCALE/(1<<15)), sp->encode_meth) & 0xff);
- luv3 += 3;
- }
- }
- static void
- _logLuvNop(LogLuvState* sp, tidata_t op, int n)
- {
- (void) sp; (void) op; (void) n;
- }
- static int
- LogL16GuessDataFmt(TIFFDirectory *td)
- {
- #define PACK(s,b,f) (((b)<<6)|((s)<<3)|(f))
- switch (PACK(td->td_samplesperpixel, td->td_bitspersample, td->td_sampleformat)) {
- case PACK(1, 32, SAMPLEFORMAT_IEEEFP):
- return (SGILOGDATAFMT_FLOAT);
- case PACK(1, 16, SAMPLEFORMAT_VOID):
- case PACK(1, 16, SAMPLEFORMAT_INT):
- case PACK(1, 16, SAMPLEFORMAT_UINT):
- return (SGILOGDATAFMT_16BIT);
- case PACK(1, 8, SAMPLEFORMAT_VOID):
- case PACK(1, 8, SAMPLEFORMAT_UINT):
- return (SGILOGDATAFMT_8BIT);
- }
- #undef PACK
- return (SGILOGDATAFMT_UNKNOWN);
- }
- static uint32
- multiply(size_t m1, size_t m2)
- {
- uint32 bytes = m1 * m2;
- if (m1 && bytes / m1 != m2)
- bytes = 0;
- return bytes;
- }
- static int
- LogL16InitState(TIFF* tif)
- {
- TIFFDirectory *td = &tif->tif_dir;
- LogLuvState* sp = DecoderState(tif);
- static const char module[] = "LogL16InitState";
- assert(sp != NULL);
- assert(td->td_photometric == PHOTOMETRIC_LOGL);
- /* for some reason, we can't do this in TIFFInitLogL16 */
- if (sp->user_datafmt == SGILOGDATAFMT_UNKNOWN)
- sp->user_datafmt = LogL16GuessDataFmt(td);
- switch (sp->user_datafmt) {
- case SGILOGDATAFMT_FLOAT:
- sp->pixel_size = sizeof (float);
- break;
- case SGILOGDATAFMT_16BIT:
- sp->pixel_size = sizeof (int16);
- break;
- case SGILOGDATAFMT_8BIT:
- sp->pixel_size = sizeof (uint8);
- break;
- default:
- TIFFErrorExt(tif->tif_clientdata, tif->tif_name,
- "No support for converting user data format to LogL");
- return (0);
- }
- sp->tbuflen = multiply(td->td_imagewidth, td->td_rowsperstrip);
- if (multiply(sp->tbuflen, sizeof (int16)) == 0 ||
- (sp->tbuf = (tidata_t*) _TIFFmalloc(sp->tbuflen * sizeof (int16))) == NULL) {
- TIFFErrorExt(tif->tif_clientdata, module, "%s: No space for SGILog translation buffer",
- tif->tif_name);
- return (0);
- }
- return (1);
- }
- static int
- LogLuvGuessDataFmt(TIFFDirectory *td)
- {
- int guess;
- /*
- * If the user didn't tell us their datafmt,
- * take our best guess from the bitspersample.
- */
- #define PACK(a,b) (((a)<<3)|(b))
- switch (PACK(td->td_bitspersample, td->td_sampleformat)) {
- case PACK(32, SAMPLEFORMAT_IEEEFP):
- guess = SGILOGDATAFMT_FLOAT;
- break;
- case PACK(32, SAMPLEFORMAT_VOID):
- case PACK(32, SAMPLEFORMAT_UINT):
- case PACK(32, SAMPLEFORMAT_INT):
- guess = SGILOGDATAFMT_RAW;
- break;
- case PACK(16, SAMPLEFORMAT_VOID):
- case PACK(16, SAMPLEFORMAT_INT):
- case PACK(16, SAMPLEFORMAT_UINT):
- guess = SGILOGDATAFMT_16BIT;
- break;
- case PACK( 8, SAMPLEFORMAT_VOID):
- case PACK( 8, SAMPLEFORMAT_UINT):
- guess = SGILOGDATAFMT_8BIT;
- break;
- default:
- guess = SGILOGDATAFMT_UNKNOWN;
- break;
- #undef PACK
- }
- /*
- * Double-check samples per pixel.
- */
- switch (td->td_samplesperpixel) {
- case 1:
- if (guess != SGILOGDATAFMT_RAW)
- guess = SGILOGDATAFMT_UNKNOWN;
- break;
- case 3:
- if (guess == SGILOGDATAFMT_RAW)
- guess = SGILOGDATAFMT_UNKNOWN;
- break;
- default:
- guess = SGILOGDATAFMT_UNKNOWN;
- break;
- }
- return (guess);
- }
- static int
- LogLuvInitState(TIFF* tif)
- {
- TIFFDirectory* td = &tif->tif_dir;
- LogLuvState* sp = DecoderState(tif);
- static const char module[] = "LogLuvInitState";
- assert(sp != NULL);
- assert(td->td_photometric == PHOTOMETRIC_LOGLUV);
- /* for some reason, we can't do this in TIFFInitLogLuv */
- if (td->td_planarconfig != PLANARCONFIG_CONTIG) {
- TIFFErrorExt(tif->tif_clientdata, module,
- "SGILog compression cannot handle non-contiguous data");
- return (0);
- }
- if (sp->user_datafmt == SGILOGDATAFMT_UNKNOWN)
- sp->user_datafmt = LogLuvGuessDataFmt(td);
- switch (sp->user_datafmt) {
- case SGILOGDATAFMT_FLOAT:
- sp->pixel_size = 3*sizeof (float);
- break;
- case SGILOGDATAFMT_16BIT:
- sp->pixel_size = 3*sizeof (int16);
- break;
- case SGILOGDATAFMT_RAW:
- sp->pixel_size = sizeof (uint32);
- break;
- case SGILOGDATAFMT_8BIT:
- sp->pixel_size = 3*sizeof (uint8);
- break;
- default:
- TIFFErrorExt(tif->tif_clientdata, tif->tif_name,
- "No support for converting user data format to LogLuv");
- return (0);
- }
- sp->tbuflen = multiply(td->td_imagewidth, td->td_rowsperstrip);
- if (multiply(sp->tbuflen, sizeof (uint32)) == 0 ||
- (sp->tbuf = (tidata_t*) _TIFFmalloc(sp->tbuflen * sizeof (uint32))) == NULL) {
- TIFFErrorExt(tif->tif_clientdata, module, "%s: No space for SGILog translation buffer",
- tif->tif_name);
- return (0);
- }
- return (1);
- }
- static int
- LogLuvSetupDecode(TIFF* tif)
- {
- LogLuvState* sp = DecoderState(tif);
- TIFFDirectory* td = &tif->tif_dir;
- tif->tif_postdecode = _TIFFNoPostDecode;
- switch (td->td_photometric) {
- case PHOTOMETRIC_LOGLUV:
- if (!LogLuvInitState(tif))
- break;
- if (td->td_compression == COMPRESSION_SGILOG24) {
- tif->tif_decoderow = LogLuvDecode24;
- switch (sp->user_datafmt) {
- case SGILOGDATAFMT_FLOAT:
- sp->tfunc = Luv24toXYZ;
- break;
- case SGILOGDATAFMT_16BIT:
- sp->tfunc = Luv24toLuv48;
- break;
- case SGILOGDATAFMT_8BIT:
- sp->tfunc = Luv24toRGB;
- break;
- }
- } else {
- tif->tif_decoderow = LogLuvDecode32;
- switch (sp->user_datafmt) {
- case SGILOGDATAFMT_FLOAT:
- sp->tfunc = Luv32toXYZ;
- break;
- case SGILOGDATAFMT_16BIT:
- sp->tfunc = Luv32toLuv48;
- break;
- case SGILOGDATAFMT_8BIT:
- sp->tfunc = Luv32toRGB;
- break;
- }
- }
- return (1);
- case PHOTOMETRIC_LOGL:
- if (!LogL16InitState(tif))
- break;
- tif->tif_decoderow = LogL16Decode;
- switch (sp->user_datafmt) {
- case SGILOGDATAFMT_FLOAT:
- sp->tfunc = L16toY;
- break;
- case SGILOGDATAFMT_8BIT:
- sp->tfunc = L16toGry;
- break;
- }
- return (1);
- default:
- TIFFErrorExt(tif->tif_clientdata, tif->tif_name,
- "Inappropriate photometric interpretation %d for SGILog compression; %s",
- td->td_photometric, "must be either LogLUV or LogL");
- break;
- }
- return (0);
- }
- static int
- LogLuvSetupEncode(TIFF* tif)
- {
- LogLuvState* sp = EncoderState(tif);
- TIFFDirectory* td = &tif->tif_dir;
- switch (td->td_photometric) {
- case PHOTOMETRIC_LOGLUV:
- if (!LogLuvInitState(tif))
- break;
- if (td->td_compression == COMPRESSION_SGILOG24) {
- tif->tif_encoderow = LogLuvEncode24;
- switch (sp->user_datafmt) {
- case SGILOGDATAFMT_FLOAT:
- sp->tfunc = Luv24fromXYZ;
- break;
- case SGILOGDATAFMT_16BIT:
- sp->tfunc = Luv24fromLuv48;
- break;
- case SGILOGDATAFMT_RAW:
- break;
- default:
- goto notsupported;
- }
- } else {
- tif->tif_encoderow = LogLuvEncode32;
- switch (sp->user_datafmt) {
- case SGILOGDATAFMT_FLOAT:
- sp->tfunc = Luv32fromXYZ;
- break;
- case SGILOGDATAFMT_16BIT:
- sp->tfunc = Luv32fromLuv48;
- break;
- case SGILOGDATAFMT_RAW:
- break;
- default:
- goto notsupported;
- }
- }
- break;
- case PHOTOMETRIC_LOGL:
- if (!LogL16InitState(tif))
- break;
- tif->tif_encoderow = LogL16Encode;
- switch (sp->user_datafmt) {
- case SGILOGDATAFMT_FLOAT:
- sp->tfunc = L16fromY;
- break;
- case SGILOGDATAFMT_16BIT:
- break;
- default:
- goto notsupported;
- }
- break;
- default:
- TIFFErrorExt(tif->tif_clientdata, tif->tif_name,
- "Inappropriate photometric interpretation %d for SGILog compression; %s",
- td->td_photometric, "must be either LogLUV or LogL");
- break;
- }
- return (1);
- notsupported:
- TIFFErrorExt(tif->tif_clientdata, tif->tif_name,
- "SGILog compression supported only for %s, or raw data",
- td->td_photometric == PHOTOMETRIC_LOGL ? "Y, L" : "XYZ, Luv");
- return (0);
- }
- static void
- LogLuvClose(TIFF* tif)
- {
- TIFFDirectory *td = &tif->tif_dir;
- /*
- * For consistency, we always want to write out the same
- * bitspersample and sampleformat for our TIFF file,
- * regardless of the data format being used by the application.
- * Since this routine is called after tags have been set but
- * before they have been recorded in the file, we reset them here.
- */
- td->td_samplesperpixel =
- (td->td_photometric == PHOTOMETRIC_LOGL) ? 1 : 3;
- td->td_bitspersample = 16;
- td->td_sampleformat = SAMPLEFORMAT_INT;
- }
- static void
- LogLuvCleanup(TIFF* tif)
- {
- LogLuvState* sp = (LogLuvState *)tif->tif_data;
- assert(sp != 0);
- tif->tif_tagmethods.vgetfield = sp->vgetparent;
- tif->tif_tagmethods.vsetfield = sp->vsetparent;
- if (sp->tbuf)
- _TIFFfree(sp->tbuf);
- _TIFFfree(sp);
- tif->tif_data = NULL;
- _TIFFSetDefaultCompressionState(tif);
- }
- static int
- LogLuvVSetField(TIFF* tif, ttag_t tag, va_list ap)
- {
- LogLuvState* sp = DecoderState(tif);
- int bps, fmt;
- switch (tag) {
- case TIFFTAG_SGILOGDATAFMT:
- 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 SGILOGDATAFMT_FLOAT:
- bps = 32, fmt = SAMPLEFORMAT_IEEEFP;
- break;
- case SGILOGDATAFMT_16BIT:
- bps = 16, fmt = SAMPLEFORMAT_INT;
- break;
- case SGILOGDATAFMT_RAW:
- bps = 32, fmt = SAMPLEFORMAT_UINT;
- TIFFSetField(tif, TIFFTAG_SAMPLESPERPIXEL, 1);
- break;
- case SGILOGDATAFMT_8BIT:
- bps = 8, fmt = SAMPLEFORMAT_UINT;
- break;
- default:
- TIFFErrorExt(tif->tif_clientdata, tif->tif_name,
- "Unknown data format %d for LogLuv compression",
- sp->user_datafmt);
- return (0);
- }
- TIFFSetField(tif, TIFFTAG_BITSPERSAMPLE, bps);
- TIFFSetField(tif, TIFFTAG_SAMPLEFORMAT, fmt);
- /*
- * Must recalculate sizes should bits/sample change.
- */
- tif->tif_tilesize = isTiled(tif) ? TIFFTileSize(tif) : (tsize_t) -1;
- tif->tif_scanlinesize = TIFFScanlineSize(tif);
- return (1);
- case TIFFTAG_SGILOGENCODE:
- sp->encode_meth = va_arg(ap, int);
- if (sp->encode_meth != SGILOGENCODE_NODITHER &&
- sp->encode_meth != SGILOGENCODE_RANDITHER) {
- TIFFErrorExt(tif->tif_clientdata, tif->tif_name,
- "Unknown encoding %d for LogLuv compression",
- sp->encode_meth);
- return (0);
- }
- return (1);
- default:
- return (*sp->vsetparent)(tif, tag, ap);
- }
- }
- static int
- LogLuvVGetField(TIFF* tif, ttag_t tag, va_list ap)
- {
- LogLuvState *sp = (LogLuvState *)tif->tif_data;
- switch (tag) {
- case TIFFTAG_SGILOGDATAFMT:
- *va_arg(ap, int*) = sp->user_datafmt;
- return (1);
- default:
- return (*sp->vgetparent)(tif, tag, ap);
- }
- }
- static const TIFFFieldInfo LogLuvFieldInfo[] = {
- { TIFFTAG_SGILOGDATAFMT, 0, 0, TIFF_SHORT, FIELD_PSEUDO,
- TRUE, FALSE, "SGILogDataFmt"},
- { TIFFTAG_SGILOGENCODE, 0, 0, TIFF_SHORT, FIELD_PSEUDO,
- TRUE, FALSE, "SGILogEncode"}
- };
- int
- TIFFInitSGILog(TIFF* tif, int scheme)
- {
- static const char module[] = "TIFFInitSGILog";
- LogLuvState* sp;
- assert(scheme == COMPRESSION_SGILOG24 || scheme == COMPRESSION_SGILOG);
- /*
- * Allocate state block so tag methods have storage to record values.
- */
- tif->tif_data = (tidata_t) _TIFFmalloc(sizeof (LogLuvState));
- if (tif->tif_data == NULL)
- goto bad;
- sp = (LogLuvState*) tif->tif_data;
- _TIFFmemset((tdata_t)sp, 0, sizeof (*sp));
- sp->user_datafmt = SGILOGDATAFMT_UNKNOWN;
- sp->encode_meth = (scheme == COMPRESSION_SGILOG24) ?
- SGILOGENCODE_RANDITHER : SGILOGENCODE_NODITHER;
- sp->tfunc = _logLuvNop;
- /*
- * Install codec methods.
- * NB: tif_decoderow & tif_encoderow are filled
- * in at setup time.
- */
- tif->tif_setupdecode = LogLuvSetupDecode;
- tif->tif_decodestrip = LogLuvDecodeStrip;
- tif->tif_decodetile = LogLuvDecodeTile;
- tif->tif_setupencode = LogLuvSetupEncode;
- tif->tif_encodestrip = LogLuvEncodeStrip;
- tif->tif_encodetile = LogLuvEncodeTile;
- tif->tif_close = LogLuvClose;
- tif->tif_cleanup = LogLuvCleanup;
- /* override SetField so we can handle our private pseudo-tag */
- _TIFFMergeFieldInfo(tif, LogLuvFieldInfo,
- TIFFArrayCount(LogLuvFieldInfo));
- sp->vgetparent = tif->tif_tagmethods.vgetfield;
- tif->tif_tagmethods.vgetfield = LogLuvVGetField; /* hook for codec tags */
- sp->vsetparent = tif->tif_tagmethods.vsetfield;
- tif->tif_tagmethods.vsetfield = LogLuvVSetField; /* hook for codec tags */
- return (1);
- bad:
- TIFFErrorExt(tif->tif_clientdata, module,
- "%s: No space for LogLuv state block", tif->tif_name);
- return (0);
- }
- #endif /* LOGLUV_SUPPORT */
- /* vim: set ts=8 sts=8 sw=8 noet: */